upwork__agency_rooms
ChatGPTAgency-only writes to message room settings and membership. Does not list rooms, read messages, or send chat (get_messages / send_message). Obtain room_id from get_messages list_rooms or find_room. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - update: Update room settings. Params: input (object, required — RoomUpdateInputV2). - add_member: Add user to a room. Params: room_id (string, required), users (array of RoomUserInput, required). - remove_member: Remove user from a room. Params: room_id (string, required), user_id (string, required), org_id (string, required).
upwork__boost_profile
ChatGPTManage freelancer profile visibility boosters: availability badge (PIB) and profile boost ad. All write operations spend Connects and WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a draft for review — call confirm_draft to execute after explicit user approval. Actions: - get_status: Read current state of visibility boosters: availability badge (on/off, weekly Connects budget, cost) and profile boost ad (status, bid, targeting). No params required. - toggle_availability_badge: Enable or disable the Availability Badge (Available Now). Spending up to max_weekly_connects Connects per week when enabled. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Params: enabled (bool, required), max_weekly_connects (int, required when enabling — maximum Connects to spend per week, e.g. 8). - get_suggested_boost_bid: Get the suggested Connects bid per click for a profile boost. Params: l1_category_id (string, required — L1 category ID), l3_category_id (string, required — L3 specialty ID). - create_profile_boost: Create a profile boost ad to appear at the top of search results. Charged only when a client clicks. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Params: bid_connects (int, required — Connects per click), l1_category_id (string, required), l3_category_id (string, required), daily_budget (int, optional — max Connects/day), total_budget (int, optional — lifetime Connects cap), end_date (string, optional — yyyy-mm-dd). - manage_profile_boost: Stop, pause, restart, or update an existing profile boost ad. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Params: action (string, required — one of: stop, pause, restart, update), ad_id (string, required), bid_connects (int, optional — for update), daily_budget (int, optional — for update), total_budget (int, optional — for update), end_date (string, optional — yyyy-mm-dd, for update).
upwork__confirm_attachment_upload
ChatGPTConfirm uploaded files so they are retained in Upwork storage. Uses the same attachment backends as start_attachment_upload. Requires the upload context and the file_uid values returned from upload. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - confirm: Params: file_ids (array of strings, required — file_uid values to confirm); task_id (string, required — the task_id returned by start_attachment_upload for these files); context (string, optional — taken from the upload session; if given it must name the same backend the session was started with, or the call is refused). Only file_uid values that get_upload_status reports as done for that task_id can be confirmed; anything else is refused. Upload sessions last 30 minutes, so confirm promptly. Files uploaded through the inline component are already stored and must NOT be confirmed.
upwork__confirm_draft
ChatGPTConfirm and execute a previously prepared draft for job postings, proposals, invitation accepts, invitations, milestones, profile updates, and offer decline / request-changes. Requires draft_id from the prior draft action. Do NOT use to ACCEPT an offer — accepting is finalized on upwork.com via the finalize_url returned by respond_to_offer accept. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Each write requires separate confirmation even if the user said "approve all". Actions: - confirm: Confirm and execute a draft using draft_id only. Params: type (string, required), draft_id (string, required — from the create draft response). Do not pass the returned params; they are loaded from server storage.
upwork__confirm_preview
ChatGPTConfirm and execute a previously prepared preview for job postings, proposals, invitation accepts, invitations, milestones, profile updates, and offer decline / request-changes. Requires preview_id from the prior preview action. Do NOT use to ACCEPT an offer — accepting is finalized on upwork.com via the finalize_url returned by respond_to_offer accept. For job postings specifically: calling create_draft (type='job_posting') always saves a real DRAFT — it never publishes, no matter what the action is labeled, and it leaves no preview to confirm here. To publish an existing job posting, call post_job update with status: 'published' and confirm that (type='job_update') — that step makes the job live to whoever its access level allows, opening it for proposals immediately. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - confirm: Confirm and execute a preview using preview_id only. Params: type (string, required), preview_id (string, required — from the create preview response; the legacy name draft_id is also accepted). Do not pass the returned params; they are loaded from server storage.
upwork__end_contract
ChatGPTEnd a contract with a reason code, optionally attaching a freelancer review (six 1–5 scores + a comment). Runs through MCP (draft -> confirm_preview). Use list_contracts action=end_reasons to look up reason codes first. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview — call confirm_preview only after explicit user approval. Actions: - end: Prepare a contract-end draft. Params: contract_id (string, required), reason_id (string, required). Optional freelancer review: quality_score, skills_score, communication_score, availability_score, deadlines_score, cooperation_score (integers 1–5) and comment (string). Provide all six scores together for a complete review.
upwork__find_freelancers
ChatGPTFreelancers on the marketplace: search them by keywords and filters, get the freelancers recommended for one of your job postings (smart_search), or view one profile. A result whose rate carries "*Boosted" is a paid ad placement the freelancer bought, not a ranking of merit — say so when presenting it. Actions: search, get_profile, smart_search. Call get_tool_help(tool_name="find_freelancers") for how to use this tool.
upwork__find_jobs
ChatGPTSearch and view marketplace job postings. Actions: - search: Search for jobs matching filters. Params: query (string), job_type (fixed/hourly), experience_level (entry_level/intermediate/expert), budget_min (number), budget_max (number) (each must be greater than 0; budget_min cannot exceed budget_max; omit a bound to leave it open), workload (full_time/part_time/as_needed), verified_payment_only (boolean — only jobs from clients with a verified payment method), proposals_min (number), proposals_max (number) (filter by number of proposals already submitted; e.g. proposals_max=5 for jobs with fewer than 5 proposals), client_hires_min (number), client_hires_max (number) (filter by the client's past hire count; e.g. client_hires_min=0, client_hires_max=0 for clients with no hires, or client_hires_min=10 for 10+). Range bounds must be >= 0 and min cannot exceed max; omit a bound to leave it open. sort (recency/relevance/client_total_charge/client_rating — how to order results; omit for the default recency order), previous_clients_only (boolean — only jobs from clients you have worked with before), timezone (string — filter by the client's timezone), location (string — filter by the client's location), limit (integer, 1–10, default 10), cursor (string — omit for the first page; to fetch the next page, repeat the SAME filters with cursor set to the endCursor from the previous response's pageInfo, and only when pageInfo.hasNextPage is true). A client's preferred qualifications are NOT in search results — use get on a specific job to see them. Each result includes proposal_count (proposals submitted so far). Hire/invite liveness (total_hired, invites_sent) is NOT in search — use get on a specific job before assuming it is still open. There is NO created_after/date filter — the signed-in marketplace search API exposes none (the public search has days_posted, but find_jobs uses the signed-in search for sort and extra filters). For the freshest jobs use sort=recency (already the default order); each result carries created_date/published_date for client-side narrowing to a time window. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - get: Get full details of a job posting including client company info, work history, connects, and the client's preferred qualifications (returned as preferred_qualifications: e.g. contractor_type, minimum job success score, English level, rising_talent, earnings). When present, surface the preferred qualifications to the user so they know what the client is looking for. The response includes connects_cost (the Connects required to apply), connects_balance, and can_apply when available. Params: id or job_id (string, required — the numeric id from find_jobs action=search, a ~02… ciphertext, or a full Upwork job URL, e.g. https://www.upwork.com/jobs/~0212345 — all accepted directly, no need to extract or reformat the id yourself). If this returns access denied for a marketplace job ID, use get_job_posting action=get_marketplace with the same id. - smart_search: Auto-fetch your profile skills and search matching jobs. Params: cursor (string, optional — omit for the first page; to fetch the next page repeat with cursor set to the endCursor from the previous response's pageInfo, only when pageInfo.hasNextPage is true).
upwork__find_saved_jobs
ChatGPTBrowse your saved (favorite) job postings. Each saved job includes a ready-to-use url. Use find_jobs action=get for full job details. Use save_job action=save or action=unsave to update bookmarks. Actions: - list: List saved jobs for the current freelancer. Params: limit (integer, 1–10, default 10), offset (number, default 0), job_ids (array of string, optional — filter by job IDs), followed (boolean, optional — defaults to true, i.e. jobs the user currently has saved. Unsaving keeps the record with followed=false, so followed=false lists jobs that were saved and then removed — not current bookmarks). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - count: Get the number of saved jobs. No params.
upwork__get_account
ChatGPTView account info, organization details, teams and the member roster. Actions: - get_user_details: Get user details. Params: id (string, optional — defaults to current user). - get_organization: Get current organization, including its teams (child organizations) when it has any. Teams are informational: they are NOT selectable accounts, so a team id must never be passed as org_uid — list_accounts returns the only usable org_uid values. No params. - list_teams: List the teams you belong to under the selected organization, optionally filtered by what you may do in each. Params: action (string, optional — one of approve_milestone, end_contract, hire, invite_freelancer, post_job, send_offer); with it each team carries allowed true/false. Team ids are team_uid FILTER values for other tools — never an org_uid. whole_organization describes acting company-wide, which is done by omitting team_uid, not by passing the org uid. - list_members: List the people on the selected client account: abbreviated name, Owner or Member, Active/Inactive, join date, company-wide roles and the teams they sit in with their roles there; an owner is listed once. Params: limit (integer, optional), offset (integer, optional — next_offset from the previous page). - get_company: Get the company details of the organization you are acting as. To read a different account of yours, call again with that org_uid. For a marketplace job's client, use find_jobs action=get with the job ID instead. No params.
upwork__get_agency
ChatGPTView your agency: teams, members, portfolio projects, and organization details. To change the portfolio, use update_agency. Use get_messages list_rooms to view rooms. Actions: - list_agencies: List agencies you belong to. No params required. - list_teams: List teams (child organizations) in the current org. Params: limit (integer, 1–10, default 10). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - list_members: List staff members in the current org. Params: limit (integer, 1–10, default 10), cursor (string, optional — pageInfo.endCursor from a prior list for the next page; alias: after). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - list_projects: List this agency's portfolio projects — the showcase clients read when deciding whether to engage the agency. Returns title, description, project URL, completion date, rank and skill ids. Params: limit (integer, 1-20), offset (integer, optional — for the next page; total_count and has_more are in the response). Project attachments are not exposed yet. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - get_profile: View your agency's public profile: summary, website, size, minimum project size, year founded, office locations and skills. Reads the agency for the selected account, so no params are required. Rates and services are not included yet. Street addresses are never returned — locations give city, state, country and postal code.
upwork__get_agency_dashboard
ChatGPTGet a consolidated overview of what's new: invitations, offers, messages, matching jobs, contracts, plus agency teams and members — all in one call. Access is scoped by your agency role. Actions: - check: Check all activity sources in parallel. No params required.
upwork__get_client_dashboard
ChatGPTGet a consolidated overview of what's new: proposals received (grouped by job), pending offers, messages, and contract updates — all in one call. Actions: - check: Check all activity sources in parallel. No params required. proposals.jobs lists the client's open postings, newest first; each carries its status, created_time and applicant funnel (applicants, new_applicants, shortlisted, messaged, offered, hired, pending_invitations). The newest 5 proposal rows (with proposals_total) are read for at most 5 postings; the rest carry proposals_omitted. new_ai_interviews: unopened rows with AI interview results ready.
upwork__get_client_financials
ChatGPTView client financial reports: timesheets, work diaries, work-diary snapshots, and transaction history. timesheets organization_id defaults to org_uid. transactions may require elevated org permissions. Actions: - summary: Get the client Weekly financial summary for one ISO week (mirrors Reports → Weekly summary): total charged and hours, the hourly/fixed breakdown, and the Top-5 contracts and Top-5 activities by charge. All charged amounts in this summary are USD, reported by the response-level currency field. Params: from_date (string, yyyy-mm-dd, optional — selects the week containing this date; defaults to the current week). The summary is a single Monday–Sunday week. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - timesheets: Get timesheets (hours logged on your contracts). Params: organization_id (string — defaults to org_uid), from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd, max 53 weeks from from_date) — at least one date required; when only from_date is set, to_date defaults to today, contract_ids (array of string), limit (number). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - work_diary: Get work diaries (time cells: activity level, screenshots, keyboard/mouse activity). With NO contract_id it returns all work diaries across the company for a single day — the marketplace 'All work diaries' view. Pass contract_id to drill into one contract's diary. Params: contract_id (string, optional — a contract id from list_contracts; omit for the company-wide view), date (string, optional — yyyy-mm-dd, defaults to today UTC), activity (string, optional filter), sort_by (string, optional), paging (string, optional — company view only). - snapshots: Drill into one work-diary snapshot (a contract at a point in time): mouse/keyboard event counts, memo, task, status, and hasScreenshot — activity metadata, NOT the screenshot image. Use work_diary for the diary view; use this only for per-timestamp detail. Params: input (object, required) with contractId (string) and timestamp (integer — Unix epoch seconds, e.g. a snapshotDateTime rawValue from work_diary). - work_days: Get work days. Params: input (object, required). - transactions: Get transaction history. Params: ace_ids (array of accounting entity ID strings, optional — auto-resolved from org context if omitted), from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd) — when both dates are omitted, defaults to the last 30 days; when only from_date is set, to_date defaults to today (max 53 weeks from from_date). limit (number, default 25, max 200 — returns the most recent N transactions; total_count reports how many exist). May require elevated org permissions on some accounts. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply.
upwork__get_draft
ChatGPTDeprecated alias of get_preview (identical behaviour) — prefer get_preview. Actions: - get: Get a pending preview for review. Params: type (required — the preview type), id (required — the preview_id returned by the creating action).
upwork__get_freelancer_dashboard
ChatGPTGet a consolidated overview of what's new: invitations, offers, messages, matching jobs, contract updates, connects balance, ID verification, and promotions — all in one call. Actions: - check: Check all activity sources in parallel. No params required.
upwork__get_freelancer_financials
ChatGPTView your own financial data as a freelancer or agency: earnings and transaction history, connects balance, and time report. Use this to verify a payment or earnings release. Actions: - summary: Get your earnings summary for a period (the Weekly financial summary): per-client totals (earned, billed, fee, tax) and per-job billings, plus overall totals. This is the answer to "how much did I earn" — use transactions only for the full line-item ledger. Params: from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd) — when both omitted, defaults to the last 30 days. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - transactions: Get transaction/earnings history. Params: ace_ids (array of accounting entity ID strings, optional — auto-resolved from org context if omitted), from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd) — at least one date required. accounting_entity_id (string, optional — single ACE ID shortcut, auto-resolved if omitted). limit (number, default 25, max 200 — returns the most recent N transactions; total_count reports how many exist). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - connects_balance: Get connects balance and recent usage history. Shows current balance breakdown (free, paid, rollover) and recent connects transactions with job titles. Params: limit (integer, 1–10, default 10). - timesheets: Get your timesheets (hours logged on your contracts). Params: organization_id (string — defaults to org_uid), from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd, max 53 weeks from from_date) — at least one date required; when only from_date is set, to_date defaults to today, contract_ids (array of string), limit (integer, 1–10, default 10). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply.
upwork__get_job_posting
ChatGPTView the job postings your organization owns. Use action=list to obtain posting IDs for list_client_proposals and list_client_invitations. Prior postings are useful templates when creating new jobs. For MARKETPLACE jobs — other clients' postings — use find_jobs, which serves clients as well as freelancers: action=search to search, action=get to read one by id. Actions: - get: Get an owned job posting by ID (from get_job_posting action=list). Params: job_id or job_posting_id. For marketplace jobs use find_jobs action=get. - list: List the selected organization's owned job postings (not marketplace), paginated. Covers the whole organization, including its teams. Params: title (string, optional — partial-match filter applied upstream), team_uid (string, optional — narrow to one team; get the ids from get_account action=get_organization childOrganizations, and never pass a team id as org_uid), limit (integer, 1–10, default 10), page (integer, default 1). Returns total_count, has_more, and next_page, plus postings[] with id, title, status, access, type, created_time, updated_time, premium, and the applicant funnel (applicants, shortlisted, messaged, offered, hired, pending_invitations, new_applicants). Use the ids for list_client_proposals / list_client_invitations; to page, call again with page=next_page. Each entry is a SUMMARY, not the full posting: description, budget, skills, screening questions and invites-sent are not included — call action=get with the id when you need them. Applicant counts ARE included, so answering "how many applied/were hired" needs no follow-up call. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply.
upwork__get_messages
ChatGPTList rooms, find rooms by context, and read messages. Actions: - list_rooms: List your rooms, newest-created first and sorted by latest activity within the page — NOT a global most-recent-first inbox, and the count of rooms is not available, so never state a total. Page with cursor while hasMore is true. Params: room_type (ALL/GROUP/ONE_ON_ONE/INTERVIEW, default ALL), unread_only (boolean, optional), active_since (RFC3339 timestamp, optional — filters weakly upstream, do not rely on it to find recent conversations). unread_only=true and active_since cannot be combined. sort_order (ASC/DESC, default DESC), limit (integer, 1–100, default 10), cursor (string, optional — next_cursor from a prior list for the next page; alias: after). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - find_room: Find a room by context. Params: context_type (contract/offer/proposal/one_on_one, required), context_id (string, required), organization_id (string, required for one_on_one). For context_type=proposal: Freelancers (TALENT): cannot initiate a proposal room or send the first message on a proposal — the client must contact you first. To reply, use an existing room from list_freelancer_proposals action=get_room or get_messages action=find_room (context_type=proposal). If no room exists yet, tell the user the client needs to message first. - list_messages: List messages in a room, NEWEST FIRST — the default page is the tail of the conversation, not the whole thread. Before summarising a thread or answering whether something was ever discussed, page back with cursor while hasMore is true; the message count is not available, so never state one. Params: room_id (string, required), limit (integer, 1–100, default 20), cursor (string, optional — next_cursor from a prior call for the next page of older messages; alias: after). message_count is present only once the whole thread has been read (hasMore false); while paging, the number of messages in the room is unknown. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - get_message: Get a single message. Params: room_id (string, required), story_id (string, required).
upwork__get_preview
ChatGPTRetrieve a pending server-stored preview without consuming it. Actions: - get: Get a pending preview for review. Params: type (required — the preview type), id (required — the preview_id returned by the creating action).
upwork__get_profile
ChatGPTView freelancer profile, transaction history, and connects balance. Actions: - get: Get a freelancer profile. Params: profile_key (string, optional — starts with ~ e.g. ~01abc123, from search results). If omitted, returns the current user's own profile. Job aggregates count completed contracts: totalFixedJobs and totalHourlyJobs are the component counts, and totalJobs is their reconciled sum. - list_highlights: List your certificates and portfolio projects that can be attached to a proposal as highlights. Returns certificates (id, name) and portfolio_projects (id, title). When drafting a proposal, offer to attach the most relevant ones and pass the chosen ids to manage_proposals create as certificate_ids / portfolio_project_ids. No params. - transactions: Get transaction/earnings history. Covers the selected account's accounting entity only, so a person with more than one account sees each separately. Params: ace_ids (array of accounting entity ID strings, optional — auto-resolved from org context if omitted), from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd, max 53 weeks from from_date — split longer questions into one call per year) — at least one date required. accounting_entity_id (string, optional — single ACE ID shortcut, auto-resolved if omitted). limit (number, default 25, max 200 — returns the most recent N transactions; total_count reports how many exist). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - connects_balance: Get connects balance and recent usage history. Shows current balance breakdown (free, paid, rollover) and recent connects transactions with job titles. balance is the general Connects wallet; product_credits (when present) are ad credits spendable only on the named product and are NOT included in it — report them separately, never summed. Page with next_cursor until hasMore is false. Params: limit (integer, 1–10, default 10), cursor (string, optional — next_cursor from the prior call).
upwork__get_rate_insights
ChatGPTGet the going hourly rate range for similar jobs to guide a job's hourly budget — the same From/To range Upwork shows in the Budget step. Read-only. Use it while helping a client set an hourly budget so the suggested rate reflects the market. Actions: - get: Get the suggested hourly rate range for a job. Params: experience_level (string, required — 'entry', 'intermediate', or 'expert'); and one of occupation_id (string — from the job classification) or project_text (string — the job title/description, used to infer the occupation). Returns rate_low / rate_high (USD/hr) and based_on_jobs (sample size).
upwork__get_tool_help
ChatGPTGet the full reference for a tool: its complete description and input schema, including every action and all required and optional parameters. Call this before using a tool whose listed description is brief and you need the full field list. Params: - tool_name (string, required) — the tool to describe (a name from search_tools or the tool list).
upwork__get_upload_status
ChatGPTReturns file_uid values for files a user uploaded through the web UI. Action name is get (not get_status). Upload sessions expire 30 minutes after start_attachment_upload; poll within that window. Actions: - get: Get uploaded file metadata. Params: task_id (string, required). Returns file_uid values and file metadata. Does not return file content.
upwork__invite_freelancer
ChatGPTInvite a freelancer to a job posting. Use list_jobs to see your jobs and how many invites remain on each, then send returns a draft — call confirm_draft with type='invitation' to send. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a draft for review — call confirm_draft to execute after explicit user approval. Call get_tool_help(tool_name="invite_freelancer") for the actions, parameters, and how to use this tool.
upwork__list_accounts
ChatGPTList your available Upwork accounts (Freelancer, Client, Agency). Call this first.
upwork__list_client_invitations
ChatGPTList or get client job invitations for one owned job posting. Requires job_reference from get_job_posting action=list (postings[].id). Invitations are per-job only — there is no list_all. Actions: - list: List active invitations for a job, or get one invitation when invitation_id is set. Params: job_posting_id, job_id, or job_reference (string, required — owned posting ID from get_job_posting action=list), invitation_id (string, optional — when set, returns that invitation only; organizationId_eq comes from org_uid), limit (integer, 1–10, default 10) — list only), cursor (string, optional — pageInfo.endCursor from a prior list for the next page; alias: after; list only). Org comes from org_uid on execute_tool. - history: Declined, withdrawn, archived, rejected or invalid invitations for a job. Params: as list, plus status.
upwork__list_client_proposals
ChatGPTView proposals on your job postings. Use action=list_all to discover proposals across all postings without a job_posting_id. Use action=list when you already have job_posting_id from get_job_posting action=list. Prior proposals can provide context for drafting offers. Actions: - list: List proposals for a job, ordered as the client's own Upwork dashboard orders them. Params: job_posting_id or job_id (string, required — from get_job_posting action=list), status (string), shortlisted (bool — only proposals the client shortlisted), sort (string), search, country, top_rated, english_level, applier_type (all optional), limit (integer, 1–50, default 50), cursor (string, optional — pageInfo.endCursor from a prior list for the next page; alias: after). A proposal carrying boosted=true is one the freelancer spent their own Connects to promote to this client; its bid also carries "*Boosted". Boosted proposals lead the first page. Always tell the client which applicants paid to promote themselves — it is a paid placement, not a ranking of merit. Absence of the flag means the proposal is not boosted, which says nothing about the freelancer — a job has a limited number of paid boost slots. verified_credentials, when present, are credentials a partner institution confirmed. The response includes totalCount and pageInfo; when the user wants to review all applicants, page through with cursor until hasNextPage is false. Use list_all to discover proposals without a posting id. Declined proposals carry get_unavailable=true — that list card is the complete view; do not call get on their id. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - get: Get full proposal details. Does not work for declined proposals (see list/list_all's get_unavailable flag) — use the list card for those instead. Params: id or proposal_id (string, required). - list_all: List proposals across all your job postings. Params: status (string, optional), sort (string, optional), limit_per_job (integer, 1–10, default 10). This cross-posting scan does not flag boosted proposals — use action=list on a single posting to see which applicants promoted themselves to the client. Each job's proposals are capped at limit_per_job; a job whose pageInfo.hasNextPage is true has more applicants — page them with action=list using that job_posting_id and pageInfo.endCursor. jobs_truncated=true means the org has more postings than were scanned; narrow with action=list per posting. Declined proposals carry get_unavailable=true — that list card is the complete view; do not call get on their id. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - metadata: Get valid proposal metadata (durations, engagement values). No params needed.
upwork__list_contracts
ChatGPTView and search your contracts and time reports. ID provenance: use action=search to discover rows; for CLIENT search results use node.contract.id (not node.id) as contract_id. Offer ids for list_offers/get come from list_offers or manage_offers list, not from list_contracts search. See docs/id-chaining.md. Actions: - search: Discover contracts/offers — call this FIRST. For freelancers: vendor_id (string, defaults to your user ID). For clients: searches clientOffers automatically, across the whole organization including its teams; narrow with team_uid (string, optional — from get_account action=get_organization childOrganizations, never as org_uid). Common: contract_statuses (array: ACTIVE, CLOSED, PAUSED), start_date_before (string), end_date_after (string), limit (integer, 1–10, default 10), offset (number, default 0). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - list: Fetch known contracts by ID only — do NOT use for discovery. Params: ids (string[], required, 1–100). Obtain ids from search (node.contract.id) or get. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - get: Get full contract details by contract ID. Authoritative milestone read for active/fixed-price contracts: offer.milestones includes dueDateTime, state (plus state_label), depositAmount, fundedAmount, paid, submissionCount, and submission records at data.contractDetails.offer.milestones[].submissions[]. For manage_milestones action=reject, choose the submission with the latest submissionDateTime and pass its id — no offer-ID chaining needed. Params: contract_id (string, required). - get_original_proposal: Get the original proposal for a contract. Params: id (string, required). - timesheets: Get timesheets (time report) for contracts. Params: organization_id (string — defaults to org_uid), from_date (string, yyyy-mm-dd), to_date (string, yyyy-mm-dd, max 53 weeks from from_date) — at least one date required; when only from_date is set, to_date defaults to today, contract_ids (array of string), limit (integer, 1–10, default 10). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - end_reasons: Get valid reason codes for ending a contract. Params: reason_type (required enum: CONTRACT_END_HOURLY | CONTRACT_END_FIXED).
upwork__list_freelancer_proposals
ChatGPTView your freelancer proposals and received invitations. Prior proposals are useful examples when drafting new proposals. Note: the status 'Accepted' means the proposal was submitted and validated (NOT that the client accepted you) — each result includes a status_label with the plain-English meaning. Start with recent submitted (Accepted) proposals; when looking for the strongest templates, also check Offered/Hired examples. Actions: - list: List your proposals with status (plus status_label), bid terms, created/modified dates, job id/title, and cover letter text. Each proposal carries boosted (true/false) and, when boosted, terms.connectsBid — the Connects bid Upwork actually STORED. This is the figure to report and the way to verify a boost after submitting: it is read back from Upwork, not the number that was sent. If it differs from what the user asked for, tell them the stored value and that the bid cannot be changed through this tool. Params: status (string, default Accepted — which means 'submitted'; values: Accepted=submitted, Offered=offer received, Hired, Activated=invitation accepted, Pending=processing, Declined, Withdrawn, Archived), sort_field (CREATEDDATETIME/MODIFIEDDATETIME, default CREATEDDATETIME), sort_order (ASC/DESC, default DESC), limit (integer, 1–10, default 10), cursor (string, optional — pageInfo.endCursor from a prior list for the next page; alias: after). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - invitations: List invitations you received from clients. The 'id' field in each result is the invitation_uid needed for manage_proposals action=accept_invitation. Each result includes jobPosting (id + title); for full job details call find_jobs action=get with jobPosting.id. Params: status (string, default pending), job_posting_id (string, optional — only invitations for that job), sort_field (CREATED, default CREATED), sort_order (ASC/DESC, default ASC), limit (integer, 1–10, default 10), cursor (string, optional — pageInfo.endCursor from a prior list for the next page; alias: after). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - get: Get full details of a proposal. Each proposal carries boosted (true/false) and, when boosted, terms.connectsBid — the Connects bid Upwork actually STORED. This is the figure to report and the way to verify a boost after submitting: it is read back from Upwork, not the number that was sent. If it differs from what the user asked for, tell them the stored value and that the bid cannot be changed through this tool. Params: id or proposal_id (string, required — from list). The response may include insights: how many proposals the job has, how many the client opened, shortlisted or messaged, and what the competing applicants look like on average (earnings, completed jobs, years on Upwork, Job Success Score) plus their most common skills. Present them as the client's activity and the field so far, not as a verdict on the proposal. When insights_available is false, relay insights_note as written: never say the job has no other applicants, and never estimate the numbers yourself. - get_room: Get the messaging room for a proposal. Params: proposal_id (string, required). Freelancers (TALENT): cannot initiate a proposal room or send the first message on a proposal — the client must contact you…
upwork__list_milestones
ChatGPTView milestone state for a fixed-price contract. Returns each milestone's state (NotFunded, Active, Submitted, Paid, …), deposit/funded/paid amounts (both rawValue and displayValue are in dollars), submission count, and — for submitted milestones — the submission deliverables with downloadable links (uploaded_by_role marks FREELANCER vs CLIENT files). Use after list_contracts to check status before submit_milestones, or after the client funds/releases payment on the web. Actions: - list: List milestones for a contract. Params: contract_id (string, required — from list_contracts search/get).
upwork__list_offers
ChatGPTView and list offers. Use list_mine to see all offers for your org. Use get to fetch full details of a specific offer. The 'id' from these results is the offer_id needed for manage_proposals action=accept_invitation. Actions: - get: Get full offer details by offer ID. Milestones match list_contracts get / manage_offers get (dueDateTime, state + state_label, depositAmount, fundedAmount, paid, submissionCount). For active contracts prefer list_contracts action=get. Params: offer_id or id (string, required — from list/list_mine here; NOT clientOffers node.id from list_contracts search). - get_attachments: List the file attachments on an offer (documents the client attached to the offer), each with its file name and a downloadable link. Params: offer_id or id (string, required — the offer ID). Returns fileName and link per attachment; present each as a markdown link (the link is presigned and expires in about 15 minutes). - list: List offers for your account — offers you sent (client) or received (freelancer). No filter needed; the tool uses your org_uid automatically and covers the whole organization, including its teams. Results are newest first. Params: team_uid (string, optional — narrow to one team, from get_account action=get_organization childOrganizations; never pass a team id as org_uid). Params: limit (integer, 1–10, default 10), page (integer, default 1), Optional: states (array of strings to filter by offer state — use role-appropriate values; returned offer state matches these keys). Client: sent, awaiting_acceptance, contract_started, awaiting_approval, expired, denied, draft, withdrawn, canceled. Freelancer/agency: awaiting_your_acceptance, contract_started, awaiting_client_approval, expired, denied, draft, withdrawn, canceled. Do NOT pass searchAttribute, id, or a filter object. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - list_mine: List your offers (same as list). Results are newest first. Params: limit (integer, 1–10, default 10), page (integer, default 1), Optional: states (array of strings to filter by offer state — use role-appropriate values; returned offer state matches these keys). Client: sent, awaiting_acceptance, contract_started, awaiting_approval, expired, denied, draft, withdrawn, canceled. Freelancer/agency: awaiting_your_acceptance, contract_started, awaiting_client_approval, expired, denied, draft, withdrawn, canceled. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply.
upwork__manage_client_proposals
ChatGPTManage proposals on your job postings while reviewing applicants: shortlist/un-shortlist and decline. To HIRE an applicant, use manage_offers create_draft instead (accepting a proposal is an offer). WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - shortlist: Shortlist (or un-shortlist) a proposal on your job posting. Shortlisting flags a promising applicant for later review; it does NOT hire them (to hire, use manage_offers create_draft). Params: proposal_id (string, required), job_posting_id (string, required — numeric posting id), shortlisted (bool, optional — default true; pass false to un-shortlist). - decline: Decline a proposal on your job posting (the freelancer is notified). Unlike shortlist, this returns a preview — execute it with confirm_preview type='proposal_decline'. Params: proposal_id (string, required), job_posting_id (string, required), reason (string, required — from action=decline_reasons), message (string, optional message to applicant). - decline_reasons: List valid decline reasons for client proposals. Present them to the user and pass the chosen one to action=decline as reason.
upwork__manage_meetings
ChatGPTSchedule video meetings with the other party in a message room. One side requests a meeting of a set length and the other picks a time from the requester's bookable slots and books it. Times cannot be invented: only the slots free_slots returns can be booked. Get room_id from get_messages (list_rooms or find_room). WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview for review — call confirm_preview to execute after explicit user approval. Only one pending preview is held per action type at a time: starting this action again before the current preview is confirmed replaces it without executing it, so wait for the user's decision on the pending preview before starting another change of this type. Actions: - list: List the video meetings in a room: every booked meeting, plus the meeting request still waiting for a time if there is one. Use the pending request's event_request_id with schedule, and a booked meeting's common_id with reschedule or cancel. Each meeting carries owner: true when the CALLER booked that time and false when the other party chose it. Say which of them picked it — that is what the user wants to know about a meeting that appeared in their calendar. A booked meeting is already final for both sides. Upwork has NO ACCEPT OR DECLINE on it — its own card offers only Cancel and Reschedule — so do not ask the user to accept or approve a meeting that already exists; offer to move it or cancel it. (This is about the meeting itself. The confirm_preview step below still applies to every write this tool makes.) Params: room_id (string, required). - free_slots: Show the bookable times for the meeting requested in this room, so the user can pick one. Either party can READ these — the requester can see what they have offered, which is worth telling the user — but booking is normally the other party's step. Only bookable times are returned. Ask the user which days and times suit them first, then pass that window as from_time/to_time — unfiltered this covers the next 60 days, far too many times to read out. When the user names several separate spans ("Tuesday morning or Thursday after 2"), pass them all as windows instead of guessing one range that spans them. Converting a stated day and time into epoch milliseconds is the step that goes wrong: confirm the YEAR and apply the user's UTC offset before sending. A window already in the past is refused. Times are epoch milliseconds, which are absolute — present them in the user's own timezone with the offset stated, and read that timezone from get_account (location.timezone) rather than assuming one. The other party picks in THEIR timezone, which may differ. Pass a slot's start_time to schedule. Pass common_id instead to read where an already-booked meeting can be moved to, before reschedule. Params: room_id (string, required), common_id (string, optional — read move targets for this meeting), from_time (string, optional — epoch milliseconds; defaults to now), to_time (string, optional — epoch milliseconds; when from_time is also given it must be after it and at most 60 days after it), windows (array, optional — up to 10 {from_time, to_time} objects of epoch milliseconds, for when the user names several separate spans. A recurring preference is ONE WINDOW PER DAY, not one window: "mornings only" over a working week is five windows, one 09:00-12:00 span per day, since a single span from Monday morning to Friday noon would also match Tuesday at 3am. Ten windows is therefore about two working weeks — bound the date range with the user first rather than enumerating a month. Only slots where the whole meeting fits inside a window are returned, each window must be at least as long as the meeting, none may be in the past, and all must end within 60 days of the earliest. At most 20 slots come back). - request: Request a video meeting with the other party in a room. Returns a draft — present it and call confirm_preview with type='meeting_request' and the returned preview_id to actually send the request. THE OTHER PARTY PICKS THE TIME — that is the normal flow an…
upwork__manage_milestones
ChatGPTWRITE milestone management. create, edit, delete, and reject run through MCP (draft -> confirm_draft). fund (activate) and approve move money — funding escrow and releasing payment — so they are completed by you on Upwork and return a link, NOT a draft. No list/get actions — inspect milestone state via list_contracts action=get (offer.milestones includes state, amounts, paid, submissionCount). WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a draft for review — call confirm_draft to execute after explicit user approval. Call get_tool_help(tool_name="manage_milestones") for the actions, parameters, and how to use this tool.
upwork__manage_offers
ChatGPTCreate draft offers and manage offers to freelancers. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Creates a draft that must be finalized on upwork.com by the user. Call get_tool_help(tool_name="manage_offers") for the actions, parameters, and how to use this tool.
upwork__manage_proposals
ChatGPTSubmit, edit, or withdraw freelancer proposals. Also accept or decline client job invitations. For client offers, use respond_to_offer. Freelancers (TALENT): cannot initiate a proposal room or send the first message on a proposal — the client must contact you first. To reply, use an existing room from list_freelancer_proposals action=get_room or get_messages action=find_room (context_type=proposal). If no room exists yet, tell the user the client needs to message first. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a draft for review — call confirm_draft to execute after explicit user approval. Call get_tool_help(tool_name="manage_proposals") for the actions, parameters, and how to use this tool.
upwork__manage_talent_lists
ChatGPTSave freelancers to your talent lists ("Your Lists"), create lists, and browse the freelancers in a list. Get freelancer_id (personId) from find_freelancers. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - list_lists: List your talent lists ("Your Lists"). No params. Returns each list's list_id (only for custom lists), list_type, title, and talent_count. - create_list: Create a new talent list. Params: title (string, required), description (string, optional). - delete_list: Delete a talent list. Params: list_id (string, required — a custom list's id). - save: Save (favorite) a freelancer. Params: freelancer_id (string, required — personId from find_freelancers), list_id (string, optional — a custom list's id from list_lists). OMIT list_id to add to your default "Favorites" list, which the tool creates automatically if it doesn't exist yet. - unsave: Remove a freelancer from a list. Params: freelancer_id (string, required), list_id (string, optional — the custom list to remove from; omit to remove from your default "Favorites" list). - list_talents: List the freelancers in a list. Params: list_id (string, optional — a custom list's id; omit to list all saved freelancers across your lists), limit (integer, 1–20, default 20), offset (number, default 0). Note: newly saved freelancers may take a moment to appear here — use lists_for_talent for an immediate check. - lists_for_talent: Show which of your lists a freelancer is in. Params: freelancer_id (string, required — personId).
upwork__manage_uma_shortlist
ChatGPTUma Recruiter — the AI hiring agent on your job postings: it reviews applicants and delivers a rated shortlist with per-candidate reasons. Eligibility is per job posting, never guessed from plan tier. For manually flagging ONE applicant instead, use manage_client_proposals action=shortlist. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview — call confirm_preview only after explicit user approval. Actions: - check: Check whether Uma Recruiter can work on one of your job postings. Eligibility is decided per posting by Upwork's allocation — this reads the live answer, so do not infer it from the client's plan. Params: opening_id (string, required — numeric opening id from get_job_posting action=list). If eligible is true you may offer action=request; if the posting already has a run, action=status shows it. - request: Ask Uma Recruiter to review applicants and keep an AI shortlist for your job posting. Unlike check/status this changes state, so it returns a preview — execute it only after explicit user approval via confirm_preview type='uma_shortlist_request'. Run action=check first; if the posting is not eligible, do not offer this. If action=status already shows a run, tell the user instead of drafting. It enables shortlisting only — it creates no offers, invites or charges. Params: opening_id (string, required — numeric opening id from get_job_posting action=list). - status: How far Uma has gotten on your job posting: current phase (sourcing / rating / shortlisting / completed / stopped / ...), candidate counters, next-update estimate and a revision number that increases each time the shortlist is refreshed. Params: opening_id (string, required — numeric opening id from get_job_posting action=list). - list: The shortlist Uma built for your job posting: each candidate with Uma's rating and explanation, plus funnel stats (evaluated / recommended / shortlisted / invited). Use application_id or freelancer_id from the rows with manage_client_proposals / invite_freelancer for follow-up actions. Params: opening_id (string, required — numeric opening id from get_job_posting action=list).
upwork__post_job
ChatGPTCreate or update a job posting. For create, returns a draft for review before posting. Use confirm_draft to publish after approval. Provide skills as human-readable names (e.g., 'React', 'WordPress') — they are resolved automatically. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a draft for review — call confirm_draft to execute after explicit user approval. Call get_tool_help(tool_name="post_job") for the actions, parameters, and how to use this tool.
upwork__respond_to_offer
ChatGPTAccept, decline, or request changes to a client offer. Use this tool for all offer responses. 'decline' and 'request_changes' run through MCP (draft -> confirm_preview). 'accept' is legally binding and must be completed by you on Upwork — it returns a link. Use list_offers action=list_mine to find your offers and the offer ID. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview — call confirm_preview only after explicit user approval. Actions: - accept: Get the Upwork link to ACCEPT an offer. Accepting is legally binding (it starts the contract) and must be done by the person on Upwork — this returns a finalize_url, it does NOT accept the offer. The offer_id must come from list_offers (action=list_mine or get). Params: offer_id (string, required). - decline: Decline an offer from a client (not binding; confirm_preview type='offer_decline'). The offer_id must come from list_offers (action=list_mine or get). Params: offer_id (string, required), reason (string, optional — one of: terms, busy, duplicate, no_response, spam, error, other, requirements), message (string, optional — message to the client). - request_changes: Request changes to an offer / counter-offer (NOT binding — the client must approve and send an updated offer; confirm_preview type='offer_request_changes'). Params: offer_id (string, required — the offer to change), description (string, required — the changes you're requesting), hourly_rate (string, required — requested rate, e.g. '55.00'), currency (string, default USD), weekly_limit (number, optional), message (string, optional — note to the client).
upwork__save_job
ChatGPTManage your own marketplace job lists: save (favorite) a job to come back to it, or hide a job to keep it out of your search results. Both lists are private — the client is never notified, and hiding is not feedback about the client or a review of any kind. Hidden jobs are already excluded from future search results by the backend, so no extra filtering is needed in find_jobs. Get job_id from find_jobs (search or get). Browse saved jobs with find_saved_jobs. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - save: Save a job to your saved jobs list. Params: job_id (string, required — job posting ID). - unsave: Remove a job from your saved jobs list. Params: job_id (string, required — job posting ID). - hide: Hide a job so it stops appearing in your search results, recording why. Params: job_id (string, required — numeric id from find_jobs), reason (string, required — one of: not_interested, vague_description, unrealistic_expectations, too_many_applicants, too_old, poor_client_reviews, doesnt_match_skills, overqualified, too_low_budget, location_mismatch). The client is not notified and cannot see this. - unhide: Un-hide a job you previously hid, letting it appear in search results again. Params: job_id (string, required — numeric id from find_jobs). - hide_status: Check whether you have hidden a job, and for what reason. Params: job_id (string, required — numeric id from find_jobs).
upwork__send_message
ChatGPTCreate rooms and send messages. message is max 10240 characters — if the user's text is longer, ask them to shorten it rather than sending it truncated. Each file_attachments item needs file_id and file_name from start_attachment_upload; include image_id when present. Freelancers cannot send the first message on a proposal — see PROPOSAL MESSAGING. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: - create_room: Create a group room. Params: input (object, required — RoomCreateInputV2). - send: Send a message to a room. Params: room_id (string, required), message (string, required unless file_attachments is set). file_attachments (array, optional). - send_to_user: Send a message to a user (auto-creates 1-on-1 room if needed). Starting a conversation with a freelancer you are not yet connected to uses one of a limited number of new connections per day; the response includes remaining_connections. Optionally attach a job via job_posting_id to share it — the recipient gets a link plus a job id their assistant can open with find_jobs get. Params: user_id (string, required), org_id (string, required — recipient's org ID), job_posting_id (string, optional), message (string, required unless file_attachments is set). file_attachments (array, optional). - message_proposal: Send a message to a proposal applicant (auto-creates room if needed). Params: job_posting_id (string, required), proposal_id (string, required), message (string, required unless file_attachments is set). file_attachments (array, optional). - edit: Edit a sent message. Params: room_id (string, required), story_id (string, required), message (string, required unless file_attachments is set).
upwork__start_attachment_upload
ChatGPTRequests a secure file upload from the user. Creates a short-lived upload session and returns task_id plus fallback_url. In MCP App-capable hosts, an inline upload UI may appear; otherwise share fallback_url with the user. The inline upload UI supports up to 7 MB per file. The fallback_url page supports up to 100 MB per file and 300 MB total; use it for larger files. Never ask for base64 text or local file paths. Before calling upload, you MUST know the upload context. Ask the user to choose messages, proposals, offer, milestones, job, or invitation when it is not obvious from the conversation. Never guess a context. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. You MUST present full action details and receive explicit approval before executing. Actions: upload. Call get_tool_help(tool_name="start_attachment_upload") for the parameters and how to use this tool.
upwork__store_uploaded_files
ChatGPTApp-only tool that persists files uploaded through the inline MCP App UI. The model does not call this tool.
upwork__submit_milestones
ChatGPTSubmit milestones for client review. Use this to request approval on completed milestone work. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview — call confirm_preview only after explicit user approval. Actions: - submit: Submit a milestone for client review/approval. Params: milestone_id (string, required), message (string, required — note to the client), amount (object, optional — { decimal_amount: string, currency: string } to submit a partial amount), attachments (array of strings, optional — file uids to attach as deliverables; upload them first with the 'milestones' upload context).
upwork__update_account
ChatGPTCLIENT account only: change the company name, description, tagline, website, business location, industry, size, profile visibility and occupations freelancers see, or manage your teams — create or rename a team, invite a member by email. Not your freelancer profile (update_profile) or personal user details (get_account). Call get_account action=get_company first and pass only the fields you want to change. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview — call confirm_preview only after explicit user approval. Actions: - update_company_details: Update the company name, description, tagline and/or website freelancers see. Pass only the fields you are changing; anything omitted is left untouched. At least one is required. Params: company_name (string, optional), description (string, optional), tagline (string, optional), website (string, optional). - update_company_location: Update the company's business location. Pass only the fields you are changing; anything omitted is left untouched. At least one is required. Params: city (string, optional), state (string, optional), country_name (string, optional — full country name, e.g. 'United States'). - update_company_profile: Update the company's industry, size, profile visibility and/or occupations. Pass only the fields you are changing; anything omitted is left untouched. At least one is required. Params: industry (string, optional), size (integer, optional — employee count), visible (boolean, optional — whether the company profile is publicly visible), l3_occupations (array of string, optional — occupation identifiers). - rename_team: Rename one of the organization's teams. The company itself is not a team — rename it with update_company_details. Params: team_uid (string, required — from get_account action=list_teams), name (string, required — the new team name). - create_team: Create a new, empty team under the organization. Teams cannot be deleted afterwards, only renamed, and creating them needs a plan that includes teams (Business Plus or above; the preview refuses up front when the plan does not). Params: name (string, required — the team name, unique within the organization). - invite_member: Invite a person by email to join the organization or one of its teams; sends them an invitation email. Params: email (string, required), roles (array of strings, required — one or more of recruiter, hiring_manager, finance_admin, account_admin, work_diary, team_chat, company_chat), team_uid (string, optional — from get_account action=list_teams; omit to invite to the whole organization), message (string, optional — a personal note in the email).
upwork__update_agency
ChatGPTChange your agency: portfolio projects (add, edit, reorder, remove). Read the current state with get_agency. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview for review — call confirm_preview to execute after explicit user approval. Only one pending preview is held per action type at a time: starting this action again before the current preview is confirmed replaces it without executing it, so wait for the user's decision on the pending preview before starting another change of this type. Actions: - create_project: Add a portfolio project to this agency's showcase. Params: title (string, required), overview (string, required — what the project involved), project_url (string, optional), completion_date (string, optional — yyyy-mm-dd), skills (array of strings, optional — Upwork skill ids). Attachments are not supported yet. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - update_project: Update a portfolio project. Pass project_id plus only the fields you want to change — the tool reads the current project and merges, because the upstream call replaces the whole record and would otherwise blank anything omitted. The draft's carried_over lists exactly which fields are preserved; position is changed with reorder_projects, not here. Existing attachments cannot be preserved through this action, so the draft warns when the project has any. Params: project_id (string, required — from list_projects), title, overview, project_url, completion_date, skills (all optional). Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - reorder_projects: Reorder the portfolio. Params: project_ids (array of strings, required — EVERY project id from list_projects, in the order you want them, most prominent first). A partial list is rejected upstream, so call list_projects first and send the complete set. The draft preview shows the resulting order by title — confirm that, not the rank numbers. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply. - delete_project: Remove a portfolio project. Params: project_id (string, required — from list_projects). Deletion is permanent, and the result is verified by re-reading the portfolio because the upstream call reports acceptance rather than removal. Optional params are refinements: do not silently invent values. If the user makes a broad request, briefly surface the most relevant available refinements and proceed with only the required params plus context the user already provided. Ask before applying optional filters when the user asks for a selective result such as best, top, cheapest, near me, urgent, or only. After returning results, mention useful refinements the user can apply.
upwork__update_contract
ChatGPTUpdate contract conditions. pause and restart run through MCP (draft -> confirm_preview). Changing the weekly hours limit is completed by you on Upwork and returns a link, not a direct change. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a preview — call confirm_preview only after explicit user approval. Actions: - pause: Prepare a contract pause draft. Params: contract_id (string, required), message (string). - restart: Prepare a draft to restart a paused contract. Params: contract_id (string, required), message (string). - update_hourly_limit: Get the Upwork link to change a contract's weekly hours limit. Changing the limit notifies the freelancer and must be done by you on Upwork — this returns a finalize_url, it does NOT change the limit. Params: contract_id (string, required).
upwork__update_profile
ChatGPTManage freelancer profile: availability, employment, languages, education, and other experience. WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION WRITE OPERATION — REQUIRES EXPLICIT USER CONFIRMATION. Returns a draft for review — call confirm_draft to execute after explicit user approval. Call get_tool_help(tool_name="update_profile") for the actions, parameters, and how to use this tool.
upwork_display_freelancer_profile
ChatGPTGet a direct link to message or hire a specific freelancer on Upwork. Provide a freelancer profile ID to receive their Upwork profile URL where you can initiate contact and hiring. --- Instructions for ChatGPT: Use this tool when: - The user wants to get a link to message or hire a specific freelancer - The user asks for "the best", "top X", or recommendations after seeing search results - The user specifies freelancers they want to contact (e.g., "give me the link for John") - You've shown search results and the user wants to take action Do NOT use this tool when: - The user wants to search for freelancers (use upwork_search_freelancers for discovery/comparison) - The user asks about finding a job for themselves (this tool is for clients hiring freelancers) - The user wants to post jobs, manage contracts, or access their Upwork account --- Tool Behavior: READ-ONLY: This tool performs read-only data retrieval. It has no side effects, does not modify any data, does not access user accounts, and does not perform any actions. It only returns a URL link to a freelancer profile page. REDIRECT ONLY: This tool returns a URL that redirects the user to Upwork. All interactions (buttons, links, messages, hiring) only redirect to the Upwork environment where the user performs all actions themselves. The tool never sends messages, hires freelancers, or takes any actions on behalf of the user. --- Capability Boundary: - This tool ONLY returns a read-only URL link for a specific freelancer profile ID. No actions are performed. - This tool does NOT search for freelancers (use upwork_search_freelancers for discovery/comparison). - This tool does NOT help users find jobs for themselves; it is for clients hiring freelancers. - This tool does NOT send messages, create contracts, post jobs, hire anyone, or take any actions on the user's behalf. - This tool does NOT access, modify, or interact with the user's Upwork account in any way.
upwork_prepare_job_post
ChatGPTUse this tool when BOTH of these are true: 1. The user has a concrete hiring need with a specific role, skill, or deliverable (e.g., "Shopify developer", "logo design", "mobile app bug fixes", "10 fintech blog posts"). 2. The user either: - explicitly wants to create, draft, post, or prepare a job post, receive proposals, or structure hiring requirements into a post, OR - describes the work they need done but does NOT explicitly ask to search, find, browse, compare, shortlist, or get recommendations for freelancers directly. PREREQUISITE: A concrete hiring need must be present in the user's message. If it is not, ask: "What are you hiring for, and what should the freelancer deliver?" Then wait for the answer before calling this tool. If the user explicitly asks to search, find, browse, compare, shortlist, or get recommendations for freelancers directly, do NOT call this tool. In that case, use upwork_search_freelancers. Do NOT call this tool for: - general questions about how to hire or post on Upwork - vague requests without a specific role, skill, or deliverable - requests to browse or compare freelancers directly --- Instructions for ChatGPT: Use this tool when: - The user wants to create, draft, or prepare a job post for Upwork - The user wants to receive proposals from freelancers - The user wants help structuring their hiring requirements into a job post - The user wants to revise or refine a previously drafted job post - The user explicitly chooses the "post a job" path instead of direct freelancer search Do NOT use this tool when: - The user wants to search for, browse, or compare freelancers directly (use upwork_search_freelancers) - The user wants to contact or hire a specific freelancer (use upwork_display_freelancer_profile) - The user is asking only for general Upwork guidance without asking to draft a post - The user is asking about finding work for themselves (this tool is for clients posting jobs) --- Tool Behavior: DRAFT PREPARATION ONLY: This tool prepares a job post draft. It does not publish a job, submit a post, or perform any final action on Upwork. EDITABLE WIDGET: This tool renders an editable job post draft widget. The widget may contain fields such as title, description, and other job-post attributes. ChatGPT should prefill any fields it can reliably infer from the conversation. The user can review, edit, and complete the draft in the widget before continuing to Upwork. USE THE WIDGET AS THE PRIMARY DRAFT SURFACE: Do not treat chat as the main place to present the full draft if the widget already shows it. Use the tool to populate known fields, then let the user refine the draft in the widget or by giving follow-up instructions in chat. NO ACCOUNT ACTIONS: This tool does not access, modify, or interact with the user's Upwork account in any way. REDIRECT ONLY: The tool may return a URL that sends the user to Upwork, where they can review, edit, and complete the posting flow themselves. IMPORTANT: Do not treat this tool as a generic "help me use Upwork" tool. Use it only when preparing an actual job post draft based on the user's hiring need. --- Draft Input Rules: - The draft must reflect the user's actual hiring need, not a generic placeholder. - Do NOT use placeholder titles such as: - "Upwork Job Post" - "Need Help" - "Freelancer Needed" - Do NOT write summaries about the user wanting guidance. - The job_title and summary must describe the actual role, skill, or deliverable the user wants to hire for. Minimum content required before calling: - At least one of: - a specific role - a specific skill - a specific deliverable Strongly preferred, but not required: - project scope - timeline - experience level - relevant tools/tech stack - success criteria If some preferred details are missing, you may still call the tool once the hiring need is concrete enough to draft a useful post and prefill the widget with what is known. --- Clarification Rules: If the user's request is too vague to produce a real draft, ask a focused follow-up before calling this tool. Good follow-up questions: - "What are you hiring for?" - "What should the freelancer deliver?" - "Is this a one-time project or ongoing work?" Do NOT ask long lists of questions all at once unless necessary. Ask only for the minimum missing details needed to produce a useful draft. --- Capability Boundary: - This tool ONLY prepares a job post draft. - It does NOT publish the job. - It does NOT search for freelancers. - It does NOT return proposals or match results. - It does NOT send messages, create contracts, manage payments, or take any actions on the user's behalf. - It does NOT help users find jobs for themselves. If the user wants to browse freelancers directly instead of posting a job, use upwork_search_freelancers. If the user wants to contact a specific freelancer, use upwork_display_freelancer_profile. --- Post-Tool-Call Response Guidelines: After calling this tool, you MUST follow these rules: 1. BE CONCISE: - Keep the response short. - Do not provide a long explanation of how Upwork works. - Do not add generic hiring tips unless the user asks. - Do not reproduce the job_title or summary values 2. DO NOT IMPLY THE JOB IS POSTED: - Clearly describe the result as a draft or prepared post. - Do not say or imply that the job has been published. 3. DO NOT REPEAT DRAFT CONTENT: - Do NOT repeat the job title or summary text in chat — the widget already displays them. - You may mention the role type in a single phrase only (e.g., "a Java developer post"), never the full title or summary verbatim. 4. GUIDE THE USER TOWARD THE NEXT STEP: End your response by pointing the user to the widget and the "Finalize on Upwork.com" button. The phrasing should feel natural, not like a checklist. The user should know they can edit the draft and when to click through to Upwork — without you enumerating steps. Examples: - "Your draft is in the widget above — edit it there if anything needs adjusting, then click 'Finalize on Upwork.com' when you're ready." - "The draft is ready above. Tweak anything directly in the widget, then finalize when you're set." - "It's in the widget — edit the details there and click 'Finalize on Upwork.com' to continue." 5. AVOID GENERIC UPWORK GUIDANCE: - Do not provide generic step-by-step instructions for using Upwork unless the user explicitly asks for them. 6. DO NOT DISCUSS BUDGET OR RATES: - Do not ask about, suggest, or mention hourly rates, fixed prices, or budget in your response. The user will set those details directly on Upwork during the posting flow. Example response: "I've drafted the job post in the widget above. You can edit the title and summary directly there, then click 'Finalize on Upwork.com' when you're ready." If helpful and accurate: "I drafted a post for a Shopify developer focused on theme fixes and mobile speed. Edit the details in the widget above and click 'Finalize on Upwork.com' when ready."
upwork_search_freelancers
ChatGPTUse this tool only when BOTH of these are true: 1. The user has identified a concrete hiring need with a specific role, skill, or deliverable (e.g., "React developer", "logo designer", "video editor", "10 fintech blog posts"). 2. The user explicitly wants to search for freelancers directly — for example by asking to search, find, browse, compare, shortlist, see candidates, or get recommendations. PREREQUISITE: A concrete hiring need must be present in the user's message. If it is not, ask: "What kind of work do you need done?" and wait for the answer before calling this tool. If the user describes a concrete project but does NOT explicitly ask to search for freelancers directly — even if they say they want to "hire" someone — do NOT call this tool. In that case, use upwork_prepare_job_post. Do NOT call this tool for: - general questions about how to hire or find help - vague requests without a specific role, skill, or deliverable - concrete hiring requests that do not explicitly ask to search or browse freelancers - users looking for jobs for themselves --- Instructions for ChatGPT: Use this tool when: - The user wants to find, search a freelancer for a project or task - The user asks queries like: - "Help me find a freelancer who can [task]" - "Can you search for a freelancer who can [task]" - "Find me a [role/skill] freelancer" - "Search for freelancers who can [deliverable]" - "I'm looking for a freelancer to [task]" Do NOT use this tool when: - The user asks about finding a job for themselves - The user wants to post a job or manage job posts - The user asks to send messages, create contracts, or manage payments - The user asks to interpret images, files, or screenshots --- Tool Behavior: READ-ONLY: This tool performs read-only data retrieval from Upwork. It has no side effects, does not modify any data, does not access user accounts, and does not perform any actions on behalf of the user. It only searches and returns publicly available freelancer profile information. ALL INTERACTIONS ARE REDIRECTS: Any buttons, links, or interactive elements in the search results widget only redirect the user to the Upwork website. The tool never sends messages, hires freelancers, or performs any actions. All actual actions (hiring, messaging, viewing profiles) are performed by the user themselves in the Upwork environment after they are redirected. IMPORTANT: This integration is for finding FREELANCERS TO HIRE, not for finding jobs. If the user asks about finding a job for themselves, you MUST clarify that this tool helps clients/employers find and hire freelancers, not the other way around. Query Field Rules: - The query field MUST contain a specific skill, role, technology, or deliverable (e.g., "React developer", "logo designer", "SEO copywriter", "Shopify store setup"). - Do NOT use vague terms like "General freelancer", "any freelancer", or "help needed". - If the user has not named a skill or task, ask for it before calling this tool. --- Capability Boundary: - This tool DOES NOT help users find jobs for themselves. It is ONLY for directly searching, browsing, and comparing freelancers the user wants to review or contact. - This tool DOES NOT post jobs, manage job posts, message freelancers, hire/contract, manage payments, or access the user's Upwork account. - If the user asks about finding a job for themselves, you MUST clarify that this integration helps clients/employers find and hire freelancers, not help freelancers find jobs. - If the user wants to post a job or draft a job post, use upwork_prepare_job_post instead of this tool. - If the user expresses general hiring intent without explicit search language, prefer upwork_prepare_job_post. - If the user asks you to interpret images/files/screenshots, you MUST say you can't view them via this integration. --- Hard Constraints (Anti-Bias and Cost): - You MUST NOT use geography, nationality, region, or language variety as a proxy for price, affordability, or "cost effectiveness". - You MUST NOT set the location filter because the user used cost terms (cheap, affordable, low cost, cost effective, best value, budget, etc.). - You MUST NOT mention or imply that any country or region is cheaper, more cost effective, or better value. --- Allowed Uses of Location (Non-Cost Only): - You MAY set location ONLY if the user explicitly requests a location or region. - You MAY set location for explicit non-cost constraints stated or implied by the user, such as: - timezone coverage (for example "cover US business hours") - regulatory or compliance constraints Timezone Coverage Rule: - If the user requests timezone coverage (for example US business hours) and does not request a location, you MAY set location to a timezone-aligned set of countries to improve matching. - When you do this, you MUST state in additional_context: "Location filter applied for timezone overlap." - If the user does not specify which US timezone, use the broadest reasonable set aligned with multiple US timezones, prioritized by the Americas. --- Budget Rules: - Use hourly_rate_max ONLY if the user provides a numeric maximum (for example "under $25/hr"). Do not guess. - Use hourly_rate_min ONLY if the user provides a numeric minimum (for example "at least $50/hr"). Do not guess. - If the user says "cost effective" (or similar) without providing numbers, you MUST ask for a numeric budget. While waiting, you may run an initial search without hourly_rate_min/max. Interaction Rule: - If budget is mentioned without numbers, you MUST ask: "What is your hourly max (or monthly budget and hours/week)?" and proceed with a budget-agnostic search in parallel. --- Timezone to Location Mapping Guidance: This guidance is non-exhaustive. Do not treat any country list as the only valid set. - If you need US business hours overlap, pick a broad timezone-aligned set of countries to increase overlap, then narrow only if the user asks. - Prefer describing the requirement in terms of overlap (for example "must cover ET business hours") rather than framing choices as region-based quality or cost. - Example starting points (not exhaustive): - ET coverage: countries in the Americas with strong overlap with ET - CT coverage: countries in the Americas with strong overlap with CT - MT coverage: countries in the Americas with strong overlap with MT - PT coverage: countries in the Americas with strong overlap with PT - Any US timezone coverage: a broad Americas-focused set spanning multiple US timezones IMPORTANT: The mapping above is for timezone overlap only. It must never be framed as a cost decision. --- Post-Tool-Call Response Guidelines (Critical): After calling this tool and displaying the UI widget, you MUST follow these rules: 1. BE CONCISE: - Keep your response brief. Do NOT repeat information that is already visible in the UI widget. - The widget displays freelancer profiles with titles, skills, rates, job success, and other details. - Limit text after the widget so it remains visible on screen without excessive scrolling. 2. REFERENCE ONLY VISIBLE FREELANCERS: - Only mention or recommend freelancers that are actually shown in the UI widget. - Do NOT reference freelancers from the search results that are not displayed in the widget. - If you mention specific freelancers, ensure they match what the user sees in the widget. 3. CHAT SCOPE LIMITATION (CRITICAL): - IMPORTANT: The chat does NOT have access to the full scope of freelancer details after the search results are displayed. - You CANNOT reference a freelancer by name to ask for more information about them in subsequent messages. - If a user asks about a freelancer by name (e.g., "Tell me more about Rashida" or "What about John?"), you MUST NOT: Try to search for them again Say "I'm not seeing a freelancer named X" or ask for profile links/screenshots Attempt to provide additional details about that freelancer - Instead, you MUST immediately respond that you don't have access to individual freelancer details after the search, and direct them to click the Message button next to their profile in the search results widget above to get more information or contact that freelancer. - Example response: "I don't have access to detailed freelancer information after the search results. To learn more about [Name] or contact them, please click the Message button next to their profile in the search results widget above." 4. EXPLAIN RECOMMENDATIONS: - If you recommend a specific freelancer, clearly state WHY (e.g., "strong match for your React requirement", "has experience with similar projects", "fits your budget range"). - Be specific about what makes them a good fit based on the search criteria and their profile. 5. GUIDE TO NEXT STEP (AUTOMATICALLY USE DISPLAY PROFILE TOOL): - After showing search results, ALWAYS push the user to use the upwork_display_freelancer_profile tool to get the link to message or hire a freelancer. - The display profile tool provides a direct link to the freelancer's profile where they can message or hire them. - CRITICAL: When the user asks for "the best", "top X", "recommendations", or specifies freelancers (e.g., "give me the best 2 mobile app developers"), you MUST immediately use the display profile tool to get their message/hire links. Do NOT ask "what would you like to do" or "what's next" - just get the links. - After listing recommended freelancers, immediately call the display profile tool for each one mentioned, then provide the links. - Example flow: "Here are the top 3 matches. [Widget displays] 1) John - React expert. 2) Sarah - Node.js specialist. 3) Mike - Full-stack developer. [Then immediately use display profile tool for all 3] Here are the links to message them: [links]" - If the user asks for specific numbers (e.g., "best 2 mobile developers"), get profile links for exactly that many. - Be action-oriented: automatically use the display profile tool, don't ask permission or what they want to do next. 6. ASKING FOR MORE PROJECT INFO (FIRST SEARCH ONLY): - On the FIRST search execution, it's acceptable to ask for more project details if the query is vague or missing important information (budget, timeline, specific requirements). - However, AFTER showing search results, do NOT ask for more project info. Instead, push users to take action with the results (use display profile tool to message/hire). - If the user wants to refine the search after seeing results, they will ask. Don't proactively ask for more details after results are shown. - You may offer to help: "I can get the link to message any of these freelancers if you're interested, or we can refine the search with more specific requirements." 7. NO GENERIC UPWORK GUIDANCE:* - Do NOT provide generic instructions on how to use Upwork (e.g., "Here's how to post a job on Upwork..."). - Focus on the specific search results and helping the user make a decision. - Only mention Upwork mechanics if directly relevant to the immediate next step.
upwork_track_event
ChatGPTInternal tool: widget-only. Hidden from the model (visibility: app). Tracks SUIT analytics events from ChatGPT widgets via Snowplow.
upwork_update_job_post_draft
ChatGPTInternal tool: widget-only. Hidden from the model (visibility: app). Persists final job post draft edits and refreshes the expiry TTL before redirect to Upwork.
upwork__boost_profile
Claudeupwork__confirm_attachment_upload
Claudeupwork__confirm_draft
Claudeupwork__end_contract
Claudeupwork__find_freelancers
Claudeupwork__find_jobs
Claudeupwork__find_saved_jobs
Claudeupwork__get_account
Claudeupwork__get_agency
Claudeupwork__get_agency_dashboard
Claudeupwork__get_client_dashboard
Claudeupwork__get_client_financials
Claudeupwork__get_draft
Claudeupwork__get_freelancer_dashboard
Claudeupwork__get_freelancer_financials
Claudeupwork__get_job_posting
Claudeupwork__get_messages
Claudeupwork__get_profile
Claudeupwork__get_rate_insights
Claudeupwork__get_tool_help
Claudeupwork__get_upload_status
Claudeupwork__invite_freelancer
Claudeupwork__list_accounts
Claudeupwork__list_client_invitations
Claudeupwork__list_client_proposals
Claudeupwork__list_contracts
Claudeupwork__list_freelancer_proposals
Claudeupwork__list_milestones
Claudeupwork__list_offers
Claudeupwork__manage_client_proposals
Claudeupwork__manage_milestones
Claudeupwork__manage_offers
Claudeupwork__manage_proposals
Claudeupwork__manage_talent_lists
Claudeupwork__post_job
Claudeupwork__respond_to_offer
Claudeupwork__save_job
Claudeupwork__send_message
Claudeupwork__set_tool_mode
Claudeupwork__set_tool_permission
Claudeupwork__start_attachment_upload
Claudeupwork__store_uploaded_files
Claudeupwork__submit_milestones
Claudeupwork__update_contract
Claudeupwork__update_draft
Claudeupwork__update_profile
Claudeupwork_display_freelancer_profile
ClaudeOpen or share the Upwork profile for a specific freelancer that has already been identified, so the user can view the profile and continue on Upwork if they want to message or hire that person. --- CRITICAL: Where freelancer_id comes from - `freelancer_id` MUST be copied exactly from the `profile_id` field on an item in the `results` array returned by `upwork_search_freelancers` in this conversation (the tool’s structured output / JSON). - Do NOT invent, guess, or fabricate a value that looks like ~01…. Invalid IDs cause errors. - Do NOT use a profile_id from memory, from the web, or from a different chat. - If the user wants to open or hire someone after a search you already ran: use the `profile_id` from that search’s structured results for the person they mean. - If there has been no successful upwork_search_freelancers in this thread, run search first (or ask the user which search result they mean). Do not call this tool with a made-up ID. --- Instructions for the AI assistant: Use this tool when: - The user wants to get a link to message or hire a specific freelancer - The user specifies freelancers they want to contact (e.g., "give me the link for John") - You've shown search results and the user wants to take action Do NOT use this tool when: - The user wants to search for freelancers (use upwork_search_freelancers for discovery/comparison) - The user asks about finding a job for themselves (this tool is for clients hiring freelancers) - The user wants to post jobs, manage contracts, or access their Upwork account - You just ran upwork_search_freelancers in the same turn — always show results first and wait for the user to ask before calling this tool --- Timing Rule (all clients): NEVER call this tool in the same response turn as `upwork_search_freelancers`. Always show the search results first, then wait for the user to respond and explicitly ask for a profile link (e.g. "show me the top one", "give me the link for #2", "I want to hire the first developer"). Calling both tools in a single turn bypasses the user's ability to review results before taking action. --- Tool Behavior: READ-ONLY: This tool performs read-only data retrieval. It has no side effects, does not modify any data, does not access user accounts, and does not perform any actions. It only returns a URL link to a freelancer profile page. REDIRECT ONLY: This tool returns a URL that redirects the user to Upwork. All interactions (buttons, links, messages, hiring) only redirect to the Upwork environment where the user performs all actions themselves. The tool never sends messages, hires freelancers, or takes any actions on behalf of the user. --- Capability Boundary: - This tool ONLY returns a read-only URL link for a specific freelancer profile ID. No actions are performed. - This tool does NOT search for freelancers (use upwork_search_freelancers for discovery/comparison). - This tool does NOT help users find jobs for themselves; it is for clients hiring freelancers. - This tool does NOT send messages, create contracts, post jobs, hire anyone, or take any actions on the user's behalf. - This tool does NOT access, modify, or interact with the user's Upwork account in any way. Input Grounding Rule: - Use this tool only when a concrete freelancer_id is already available. - If the freelancer was selected from upwork_search_freelancers, the freelancer_id must be the exact profile_id from that selected search result, passed unchanged. - Never infer, fabricate, or guess a freelancer ID from a name, rank, title, or description alone. - If the user says "the second one," "that person," or uses a visible freelancer name, first resolve that reference against the latest displayed search results. If the reference cannot be resolved unambiguously, do NOT call this tool.
upwork_prepare_job_post
ClaudeDEFAULT TOOL: When the user's hiring intent is present but it is unclear whether they want to search for freelancers directly or receive proposals, prefer this tool. Use this tool when BOTH of these are true: 1. The user has a concrete hiring need with a specific role, skill, or deliverable (e.g., "Shopify developer", "logo design", "mobile app bug fixes", "10 fintech blog posts"). 2. The user either: - explicitly wants to create, draft, post, or prepare a job post, receive proposals, or structure hiring requirements into a post, OR - describes the work they need done but does NOT explicitly ask to search, find, browse, compare, shortlist, or get recommendations for freelancers directly, OR - wants to revise or refine a previously drafted job post. PREREQUISITE: A concrete hiring need must be present in the user's message. If it is not, ask: "What are you hiring for, and what should the freelancer deliver?" Then wait for the answer before calling this tool. If the user explicitly asks to search, find, browse, compare, shortlist, or get recommendations for freelancers directly, do NOT call this tool. In that case, use upwork_search_freelancers. Do NOT call this tool for: - general questions about how to hire or post on Upwork - vague requests without a specific role, skill, or deliverable - requests to browse or compare freelancers directly Tool Behavior: DRAFT PREPARATION ONLY: This tool prepares a job post draft. It does not publish a job, submit a post, or perform any final action on Upwork. EDITABLE WIDGET: This tool renders an editable job post draft widget. The widget may contain fields such as title, description, and other job-post attributes. The AI assistant should prefill any fields it can reliably infer from the conversation. The user can review, edit, and complete the draft in the widget before continuing to Upwork. USE THE WIDGET AS THE PRIMARY DRAFT SURFACE: Do not treat chat as the main place to present the full draft if the widget already shows it. Use the tool to populate known fields, then let the user refine the draft in the widget or by giving follow-up instructions in chat. ONLY APPROVED SURFACE FOR JOB POST CONTENT: This tool and its widget are the only approved way to present a job post draft to the user. Never write job post content (titles, summaries, requirements, templates, or outlines) directly in chat as a substitute for calling this tool — even when search results are poor or the user seems to need guidance. NO ACCOUNT ACTIONS: This tool does not access, modify, or interact with the user's Upwork account in any way. REDIRECT ONLY: The tool may return a URL that sends the user to Upwork, where they can review, edit, and complete the posting flow themselves. IMPORTANT: Do not treat this tool as a generic "help me use Upwork" tool. Use it only when preparing an actual job post draft based on the user's hiring need. --- Draft Input Rules: - The draft must reflect the user's actual hiring need, not a generic placeholder. - Do NOT use placeholder titles such as: - "Upwork Job Post" - "Need Help" - "Freelancer Needed" - Do NOT write summaries about the user wanting guidance. - The job_title and summary must describe the actual role, skill, or deliverable the user wants to hire for. Minimum content required before calling: - At least one of: - a specific role - a specific skill - a specific deliverable Strongly preferred, but not required: - project scope - timeline - experience level - relevant tools/tech stack - success criteria If some preferred details are missing, you may still call the tool once the hiring need is concrete enough to draft a useful post and prefill the widget with what is known. --- Clarification Rules: If the user's request is too vague to produce a real draft, ask a focused follow-up before calling this tool. Good follow-up questions: - "What are you hiring for?" - "What should the freelancer deliver?" - "Is this a one-time project or ongoing work?" Do NOT ask long lists of questions all at once unless necessary. Ask …
upwork_search_freelancers
ClaudeUse this tool only when BOTH of these are true: 1. The user has identified a concrete hiring need with a specific role, skill, or deliverable (e.g., "React developer", "logo designer", "video editor", "10 fintech blog posts"). 2. The user explicitly wants to search for freelancers directly — for example by asking to search, find, browse, compare, shortlist, see candidates, or get recommendations. If the user describes a concrete project but does NOT explicitly use search language even if they say they want to "hire" someone — do NOT call this tool. In that case, use upwork_prepare_job_post. Do NOT call this tool for: - general questions about how to hire or find help - vague requests without a specific role, skill, or deliverable - concrete hiring requests that do not explicitly ask to search or browse freelancers - users looking for jobs for themselves --- Tool Behavior: READ-ONLY: This tool performs read-only data retrieval from Upwork. It has no side effects, does not modify any data, does not access user accounts, and does not perform any actions on behalf of the user. It only searches and returns publicly available freelancer profile information. ALL INTERACTIONS ARE REDIRECTS: Any buttons, links, or interactive elements in the search results widget only redirect the user to the Upwork website. The tool never sends messages, hires freelancers, or performs any actions. All actual actions (hiring, messaging, viewing profiles) are performed by the user themselves in the Upwork environment after they are redirected. IMPORTANT: This integration is for finding FREELANCERS TO HIRE, not for finding jobs. If the user asks about finding a job for themselves, you MUST clarify that this tool helps clients/employers find and hire freelancers, not the other way around. Query Field Rules: - The query field MUST contain a specific skill, role, technology, or deliverable (e.g., "React developer", "logo designer", "SEO copywriter", "Shopify store setup"). - Do NOT use vague terms like "General freelancer", "any freelancer", or "help needed". - If the user has not named a skill or task, ask: "What skill or role are you looking to hire for?" before calling this tool. --- Capability Boundary: - This tool DOES NOT help users find jobs for themselves. It is ONLY for directly searching, browsing, and comparing freelancers the user wants to review or contact. - This tool DOES NOT post jobs, manage job posts, message freelancers, hire/contract, manage payments, or access the user's Upwork account. - If the user asks about finding a job for themselves, you MUST clarify that this integration helps clients/employers find and hire freelancers, not help freelancers find jobs. - If the user wants to post a job or draft a job post, use upwork_prepare_job_post instead of this tool. - If the user expresses general hiring intent without explicit search language, prefer upwork_prepare_job_post. - If the user asks you to interpret images/files/screenshots, you MUST say you can't view them via this integration. --- Hard Constraints (Anti-Bias and Cost): - You MUST NOT use geography, nationality, region, or language variety as a proxy for price, affordability, or "cost effectiveness". - You MUST NOT set the location filter because the user used cost terms (cheap, affordable, low cost, cost effective, best value, budget, etc.). - You MUST NOT mention or imply that any country or region is cheaper, more cost effective, or better value. --- Allowed Uses of Location (Non-Cost Only): - You MAY set location ONLY if the user explicitly requests a location or region. - You MAY set location for explicit non-cost constraints stated or implied by the user, such as: - timezone coverage (for example "cover US business hours") - regulatory or compliance constraints Timezone Coverage Rule: - If the user requests timezone coverage (for example US business hours) and does not request a location, you MAY set location to a timezone-aligned set of countries to improve matching. - When you do this, you MUST state in additional_context: …
upwork_update_job_post_draft
ClaudeInternal tool: widget-only. Hidden from the model (visibility: app). Persists final job post draft edits and refreshes the expiry TTL before redirect to Upwork.
