accept_agent_browser_policy
ChatGPTAccept a persisted preview by proposal ID, exact preview digest and expected draft etag. Applies only to this agent draft; normal agent publication activates it. Requires an organization admin and native confirmation.
apply_agent_restore
ChatGPTApplies an unexpired preview created by this same MCP connection (role_id, preview_id): a new staging draft carrying the revision's settings, voice, Playbook prompt version, products, Knowledge exclusions and presentation assets, with the restored revision as its parent, replacing the open draft the preview named. Production is unchanged: publish with publish_agent_version. Fails without changing anything when staging changed since the preview, while a publish job or a Role generation is active, or when the revision's Playbook prompt version no longer exists. Retrying returns the original result. Returns stagingRevision (the new draft), replacedDraftRevisionId, promptVersion and the staging test link. Organization admins only.
apply_coaching_changes
ChatGPTApplies selected changes of one feedback (feedback_id, changes [{id, edited_text?, edited_payload?}]) to the Role's draft, each to its own destination, and returns a per-change outcome: saved_to_draft, pending_shared_approval (a shared fact now awaiting approve_shared_fact), recapture_requested, platform_issue, unsupported, conflict or failed. Never treat the call as one result: read every change's outcome. Send the three draft tokens exactly as the last derive or list returned them (expected_revision, sections_etag, draft_etag); a draft that moved since is refused whole with draft_changed and nothing is written, so reload and retry. The response carries the draft tokens for the next apply. Nothing here reaches production: the draft is published separately. Organization admins only.
apply_custom_link_action
ChatGPTActivates, deactivates or moves up to 500 Custom Links at once (action, custom_link_ids; campaign {id} or {name} for move, created when the name is new). Returns the ids changed, and in notAppliedIds the ids that are not Custom Links of this organization. Repeating the call leaves the same end state. A deactivated link stops opening its personalized demo; a moved link's sessions, past ones included, count under its new campaign. Deactivate and move require native user confirmation.
apply_playbook_suggestions
ChatGPTApplies selected stored Playbook suggestion payloads exactly as generated. Use the preview flow instead when the wording should be edited first.
apply_playbook_update
ChatGPTApplies an unexpired preview created by this same MCP connection to the staging Playbook of the Role it was previewed for. Staging drift fails without creating a version.
approve_browser_guardrails
ChatGPTApprove a persisted restriction or instructions proposal by feedback_id/change_id. Requires native user confirmation and an organization admin/owner, including self-approval. Saves enforcement and policy-linked instructions together for new sessions across all affected agents, without publishing unrelated Playbook drafts. Revalidates scope, policy revision and evidence; replay returns the durable receipt.
approve_shared_fact
ChatGPTApproves a coached shared fact (feedback_id, change_id of a change whose outcome is pending_shared_approval and approval is null), optionally with the approver's wording (edited_text) and the question it answers (question; the statement itself when omitted). The fact becomes organization knowledge every active Role serves from the next knowledge revision, superseding an existing answer to the same question; a fact an active answer already states is approved without a new revision. Requires native user confirmation. Replay returns the recorded approval and changes nothing. Anything else is change_not_awaiting_approval. Organization admins only.
author_agent_browser_policy
ChatGPTPropose agent-specific policy changes from natural language and optional captured evidence. Searches reusable policy blocks. Generates guidance from released templates. Returns a persisted preview; never publishes.
cancel_custom_link_job
ChatGPTCancels a Custom Link job (job_id): its queued items stop, and an item already rendering finishes first. Returns the job; a finished job is returned unchanged, so repeating the call is safe.
compare_agent_revisions
ChatGPTCompares one revision of an agent (role_id, revision_id) with another of its revisions (against: a revision id) or with what staging serves now (against: "staging"). Returns from (against) and to (revision_id), each changed scope with its label and both values (before is from, after is to; scope names the get_agent_revision settings field, outcomeRubric, escalationCriteria or playbookPrompt), and playbookChanges, the Playbook sections that differ between the Playbooks the two serve; playbookTextCompared is false when either side's Playbook text could not be read, and playbookChanges then lists nothing. Reads only. Organization admins only.
create_custom_link_campaign
ChatGPTCreates a Custom Link campaign (name). When a campaign with that name already exists (case-insensitive), returns it with created false instead of failing, so a retry is safe.
create_custom_links
ChatGPTCreates 1-200 personalized demo links as one asynchronous job and returns a job id immediately — poll progress with get_custom_link_job. Before calling, confirm the batch with the user. If you retry after an error or disconnect, you MUST reuse the same operation_key — never generate a new one for the same logical batch, or duplicates may be created. Logo resolution is automatic when enabled for this environment. generate_thumbnails defaults to true; false skips preview rendering only. Supplied thumbnails are preserved. Poll get_custom_link_job with include_items: true to retrieve the finished image URLs.
delete_custom_link_campaign
ChatGPTDeletes a campaign (campaign_id). Its links are not deleted: they stay as they are but become unfiled, and their campaign attribution is lost, so campaign reports and filters no longer count their sessions under any campaign. Requires native user confirmation.
derive_coaching_changes
ChatGPTTurns one filed feedback (feedback_id) into typed, separately reviewable changes and returns them with the Role's draft tokens (draft.revision_id, sections_etag, draft_etag). Idempotent: a second call returns the first derivation with created false unless rederive is true, which supersedes that derivation's pending changes. Each change carries its kind (playbook_section, page_exclusion, voice_setting, shared_fact, recapture_request, platform_issue, screen_guidance), before/proposed content and the feedback as evidence. Carry the change ids and the three draft tokens into apply_coaching_changes. Organization admins only.
dismiss_coaching_change
ChatGPTDrops one pending change of a feedback (feedback_id, change_id), or rejects a shared fact awaiting approval, and returns the change as dismissed. Any other applied change cannot be dismissed (change_not_pending). Organization admins only.
edit_agent_guardrail_instructions
ChatGPTPreview replacement text for specific instruction IDs with their expected text hashes and draft etag. Includes read-only guidance. Keeps policy ownership and enforcement unchanged; never replaces an anonymous entire section.
file_coaching_feedback
ChatGPTFiles one piece of coaching about a Role and returns the stored row with the Role it landed on. Name exactly one target: call_id (a session; the Role is the one the session served, and screen {node_id, product_key?} and moment provenance may say which screen and turn the coaching is about, using the screenNodeId/screenProductKey from get_session_transcript) or role_id (direct feedback, no screen). kind is guidance or platform_issue. source labels who filed it for the review and defaults to platform; say widget only when relaying a note the embed collected. Send a submission_key when you may retry: a replay with the same key returns the first row with created false. Filing stores the note only; call derive_coaching_changes next. Organization admins only.
get_account
ChatGPTReturns one account rollup by registrable domain (e.g. acme.com).
get_agent_browser_policy
ChatGPTRead an initialized agent's draft browser policy, pinned reusable blocks and instruction blocks. Use instruction ID and SHA256 of current_text for targeted coaching; preserves ownership. Organization admins only.
get_agent_publish_job
ChatGPTReturns queued/running/succeeded/failed status for one agent publish job, with a safe failure message and final production link after success.
get_agent_revision
ChatGPTReturns one revision (role_id, and revision_id from list_agent_revisions) with the agent's name and id: its status (draft: the one open, editable revision, which staging serves; published: the one revision production serves; archived: published before, then replaced by a later publish), its settings as that revision holds them (identity, objective and additional instructions, capabilities, voice and speech, handoffs, opening, lead capture, launcher, Compass, entry copy, research, guidance, goals, language policy, products, demo settings, page and Knowledge exclusions), its outcome rubric and its escalation criteria (null when it has none). Identify revisions by id, never by number, which can be reused. The Playbook text is not included (playbookTextIncluded is false); get_playbook returns the current staging and production Playbooks. Comparing revisions, old Playbook text and restore wait for ENG-924. Organization admins only.
get_browser_guardrails
ChatGPTRead the policy-owned Browser guardrails Playbook section for one Role, with effective shared policies, current instructions/revisions, affected agents and classified capture choices. No credentials. Ordinary Playbook patches cannot change this section's enforcement. Organization admins only.
get_custom_link
ChatGPTReturns one Custom Link with its personalization, sharing metadata, campaign, active state, and thumbnail status.
get_custom_link_campaign_engagement
ChatGPTReturns one campaign's engagement (campaign_id): links generated, active, opened, started and engaged, leads submitted, meetings clicked and booked, and load, page-view, visitor and session counts. Production sessions only, all-time, over the links filed in the campaign now: a link moved in or out brings or takes its whole history.
get_custom_link_job
ChatGPTReturns status, progress, counters, and timestamps for an asynchronous Custom Link job (creation, thumbnail regeneration, or retry). Call this after create_custom_links to track completion; set include_items: true for per-link outcomes, logoStatus, prospectLogoUrl, and shareThumbnailUrl. Wait for a terminal job before using shareThumbnailUrl as an email image; inspect warnings for missing logos. Omit job_id to list recent jobs, newest first.
get_funnel_metrics
ChatGPTReturns aggregate funnel performance for a trailing UTC window of days. totalSessions and bookedCount count sessions; qualifiedCount and highIntentCount count unique active people. qualificationRate is the percentage of unique active people who are currently qualified, not qualified sessions divided by totalSessions. avgDurationSeconds is the mean session duration. Use for overall performance questions rather than per-session or per-person detail.
get_performance_report
ChatGPTReturns sales performance for 1-4 periods, one segment per period and environment, optionally grouped by up to two of agent, version, campaign and channel (max_groups per segment, largest first). Periods are local dates (start_date, end_date, inclusive) in timezone, an IANA name defaulting to UTC whose daylight-saving changes the day boundaries follow, or exact instants (start inclusive, end exclusive); at most 366 days each. environments defaults to production, the population get_funnel_metrics reports; add staging only when asked, and keep the environments apart. Metrics use get_funnel_metrics' definitions: sessions are analyzed sessions, booked means the analysis found a booking, and people are unique people by their current qualification and intent; every rate carries its numerator and denominator. completeness says what the metrics leave out or cannot attribute: sessions awaiting analysis or failed, sessions without an email, unknown versions, deleted agents and links; internal team and evaluation sessions stay in the metrics and are counted there. Website chats and Hobbes referral demos are never included. Each filter value reports found; an id that is not this organization's is found: false. Read context.caveats before comparing: comparisons are observational, and campaign is each link's current campaign.
get_person
ChatGPTReturns one prospect's aggregated profile by exact email address.
get_playbook
ChatGPTReturns one Role's structured staging and production Playbooks, section catalog, exact staging ETag and versions, and the staging/production links that reach that Role. Pass role_id from list_roles; omitted, the organization's default Role.
get_session
ChatGPTReturns the full sales analysis for one session: qualification status, buying intent, summary, next steps, conversation timeline, topics discussed, objections raised, booking status, and acquisition/device/location attribution. Prefer this over the transcript — it answers most questions about what happened in a session.
get_session_transcript
ChatGPTReturns the ordered conversation turns for one session. Transcripts are sensitive raw conversation data and reads count against a daily organization quota — retrieve one only when the analysis from get_session is insufficient to answer the question.
list_accounts
ChatGPTReturns B2B engagement rollups grouped by registrable email domain — a company-level view aggregating all people and sessions from that domain. Use for account- or company-level questions.
list_agent_revisions
ChatGPTReturns one agent's version history (role_id from list_roles), newest first, with the agent's name and id. Each revision is draft (the one open, editable revision, which staging serves), published (the one revision production serves) or archived (published before, then replaced by a later publish). Identify a revision by its id, never its number: a discarded or replaced draft's number is given to the next draft. limit sets how many revisions come back (default 20); hasMore says older ones exist: when it is true, call again with before_number set to the last listed number. Read one with get_agent_revision. Comparing revisions, old Playbook text and restore wait for ENG-924. Organization admins only.
list_attribution_touches
ChatGPTReturns funnel events with the attribution evidence each one captured: UTM values, which click-ID types were present (never their values), page domain and path, referrer host, and whether it is an explicit paid touch. A copied first-touch snapshot is labelled and never counted as a paid touch. Stored session attribution (needs sessions:read), People first source (needs sessions:read and people:read), and custom-link references (needs custom_links:read) are separate blocks, marked omitted_missing_scope when not granted. Page with next_cursor; every page reads the snapshot fixed by the first page.
list_coaching_changes
ChatGPTReturns the changes derived for one Role (role_id), newest first, optionally filtered by status (pending, routed, dismissed, superseded), together with where the Role's draft is now (draft.revision_id, sections_etag, draft_etag). A change marked stale was derived against an earlier draft and answers conflict if applied; re-derive its feedback. Carry the draft tokens from this response into apply_coaching_changes. Organization admins only.
list_coaching_feedback
ChatGPTReturns the coaching feedback filed against one Role (role_id from list_roles), newest first, with the Role's backlog summary (feedback_total, feedback_underived, changes_pending). Each row says whether changes were derived from it (derivation, 0 when never), how many wait for review (pending_changes) and how many a reviewer applied (routed_changes). To page, send the last row's created_at and id as before_created_at and before_id. Organization admins only.
list_custom_link_campaigns
ChatGPTReturns Custom Link campaigns and their link counts. Pass campaign_id to fetch a single campaign.
list_custom_link_sessions
ChatGPTReturns the production sessions started from one Custom Link (link_id), newest first: each session's id, lifecycle stage, opened/started/engaged/ended times and whether its analysis is available (read it with get_session). Paginated with limit and offset.
list_custom_links
ChatGPTReturns the organization's personalized demo links (Custom Links), filterable by active state, campaign, and search, including each link's current thumbnail state.
list_organizations
ChatGPTLists the organizations this verified Hobbes staff account can select, with their stable ids and names, and marks the connection's own organization. Use select_organization to act on one without reconnecting. Available only to verified Hobbes staff.
list_people
ChatGPTReturns prospect profiles aggregated across all of each person's sessions: engagement history, first/last seen, total sessions, qualification status, and buying intent. Most recently seen first. Use this for questions about people rather than individual sessions.
list_playbook_suggestions
ChatGPTLists ready Hobbes-generated suggestion groups whose changes all target Playbook sections, including stored evidence and exact proposed payloads. Suggestions exist for the default Role only; another role_id returns an empty list with unavailableReason set.
list_roles
ChatGPTReturns the organization's agent Roles (default Role first) with their ids, names, active state and objective. Every Playbook tool accepts an optional role_id from this list; a call without one acts on the default Role and its response says so.
list_sessions
ChatGPTReturns analyzed production demo sessions for the connected organization, newest first. Use the filters to narrow by outcome (qualification, buying intent, booked status) or time window before pulling individual session details.
preview_agent_browser_policy
ChatGPTPreview typed policy edits or explicit wording reconciliation for one agent draft, with an expected draft etag. Accept separately. Reuse the idempotency key when retrying the same request.
preview_agent_restore
ChatGPTPreviews restoring one published or archived revision of an agent (role_id, revision_id) to staging and stores the preview for apply_agent_restore. Returns previewId, sourceRevision, stagingRevision (what staging serves now), replacesDraft (true when staging is an open draft, which the restore deletes with its unpublished changes), each scope the restore changes (before is staging now, after is the restored revision) and playbookChanges (playbookTextCompared is false when either side's Playbook text could not be read). Nothing changes until the apply; previews expire after 24 hours. Reuse the same operation_key when retrying: the same request returns the same preview. A draft revision, or one whose Playbook prompt version no longer exists, cannot be restored, and nothing is previewed while a publish job or a Role generation is active. Organization admins only.
preview_playbook_update
ChatGPTValidates exact restricted JSON Patch operations against a cloned staging Playbook of one Role (role_id; the default Role when omitted) and returns a deterministic before/after summary. This never changes staging.
propose_browser_guardrail_instructions
ChatGPTPropose replacement wording for the policy-owned Browser guardrails Playbook section. This never adds/removes/allows a browser operation. Stores a reviewed wording-only proposal in the same ledger; approve_browser_guardrails activates a new shared policy revision with identical enforcement and the reviewed text. Preserves Role drafts; requires coaching:write and an org admin.
propose_browser_restriction
ChatGPTRequest a browser restriction in natural language without entering the coaching UI. Uses the same captured-evidence authoring service and coaching ledger. Read get_browser_guardrails first for policy/capture IDs and expected revision. Returns exact additive rules and proposed instructions, not an active policy. Requires coaching:write and an org admin; reuse operation_key to retry note submission.
publish_agent_version
ChatGPTStarts an asynchronous production publish of one Role's exact current staging version (role_id; the default Role when omitted). Requires native user confirmation and runs the existing conversation-test gate. The publish fails if the Role's draft changes after the job is queued.
regenerate_custom_link_thumbnail
ChatGPTStarts a background job that renders a new preview image for one Custom Link (link_id), replacing its current image, and returns the job; poll get_custom_link_job. Counts against the organization's render quota. Send an operation_key and reuse it when retrying: the same key returns the same job, and a key reused for a different request returns idempotency_key_reused.
rename_custom_link_campaign
ChatGPTRenames a campaign (campaign_id, name). It keeps its id and links. Another campaign with that name answers conflict; repeating the call leaves the same end state.
retry_custom_link_job
ChatGPTStarts a new background job that runs again the failed items of a Custom Link job (job_id), or the items named in item_ids (ids from itemOutcomes of get_custom_link_job), and returns it; poll get_custom_link_job. Items still queued or rendering answer conflict. Rejected items (no link was created; their input failed validation) cannot be retried: create them again with corrected input. Counts against the organization's render quota. Send an operation_key and reuse it when retrying this call: the same key returns the same job.
select_organization
ChatGPTSelects an organization from list_organizations for this conversation and returns an organization_context value. Send it as organization_context on every later Hobbes tool call in this conversation to act on that organization; calls without it act on the connection's own organization. Selecting adds no permissions: calls keep what the connection was approved for, limited by your role in that organization. It expires after 12 hours; select again when a call reports it expired. Earlier previews, jobs and confirmations stay with the organization they were made in. Available only to verified Hobbes staff.
update_custom_link
ChatGPTEdits one Custom Link (link_id): personalization, sharing copy, logo, prospect domain, active state or campaign ({id} or {name}, created when the name is new). Only the fields sent change. Changing display_name, first_name, last_name, email, company, prospect_logo_url or prospect_domain renders a new preview image as a background job, returned in job (poll get_custom_link_job), unless the link keeps a supplied share_thumbnail_url; setting share_thumbnail_url uses that image instead, and clearing it renders one. Send an operation_key and reuse it when retrying the same update. The user confirms the before and after when the client supports it; deactivating the link (active false) or sending campaign requires native user confirmation. Takes effect on the link's next open.
apply_agent_restore
Claudeapply_coaching_changes
Claudeapply_custom_link_action
Claudeapply_playbook_suggestions
Claudeapply_playbook_update
Claudeapprove_browser_guardrails
Claudeapprove_shared_fact
Claudecancel_custom_link_job
Claudecompare_agent_revisions
Claudecreate_custom_link_campaign
Claudecreate_custom_links
Claudedelete_custom_link_campaign
Claudederive_coaching_changes
Claudedismiss_coaching_change
Claudefile_coaching_feedback
Claudeget_account
Claudeget_agent_publish_job
Claudeget_agent_revision
Claudeget_browser_guardrails
Claudeget_custom_link
Claudeget_custom_link_campaign_engagement
Claudeget_custom_link_job
Claudeget_funnel_metrics
Claudeget_performance_report
Claudeget_person
Claudeget_playbook
Claudeget_session
Claudeget_session_transcript
Claudelist_accounts
Claudelist_agent_revisions
Claudelist_attribution_touches
Claudelist_coaching_changes
Claudelist_coaching_feedback
Claudelist_custom_link_campaigns
Claudelist_custom_link_sessions
Claudelist_custom_links
Claudelist_organizations
Claudelist_people
Claudelist_playbook_suggestions
Claudelist_roles
Claudelist_sessions
Claudepreview_agent_restore
Claudepreview_playbook_update
Claudepropose_browser_guardrail_instructions
Claudepropose_browser_restriction
Claudepublish_agent_version
Clauderegenerate_custom_link_thumbnail
Clauderename_custom_link_campaign
Clauderetry_custom_link_job
Claudeselect_organization
Claudeupdate_custom_link
Claude