MCP App Store
by KeenEthics
Bolta for Social Media icon

Bolta for Social Media MCP Server for ChatGPT

By Bolta Platforms, Inc.

Overview

Bolta drafts and schedules social posts for a business workspace. Write a draft, set the time it should publish, and check items waiting for approval. It uses only the social accounts already connected in that workspace, and it publishes only when you ask it to.

How to connect

  1. ChatGPT: Open the Bolta for Social Media listing in the ChatGPT app directory and choose Connect. ChatGPT listing Bolta for Social Media exposes 67 tools here.

Tools

activate-routine

ChatGPT
Enable future triggers for one existing routine. This is a fixed lifecycle operation, not a dispatcher accepting operation names or tool definitions. Arm a routine so its trigger fires. A scheduled routine starts running on its schedule from this point; a source-driven one starts reacting to events. The server re-validates the whole definition against current workspace state here, so activation can fail with a typed reason (missing source, disconnected account, plan gate) even though saving succeeded. It does not run anything now: the first run is at next_run_at (read it with get-routine). run-routine-now runs it immediately and spends credits. Also the way to resume a paused routine (same call). Tell the user what will now happen and when the first run is due.

add-posts-to-content-campaign

ChatGPT
Attach existing posts to a content campaign. Idempotent and scoped, and a PARTIAL result is the normal outcome rather than an error: a post already in the campaign comes back under already_present, and an id that is not in this workspace comes back under not_found. Both arrive as 200 alongside whatever did apply, so report what actually attached rather than assuming all of it did. Nothing here creates or publishes a post.

analyze-content-campaign-source

ChatGPT
Have Bolta read one source attached to a content campaign so its weekly writer can use it. Call it once for each source left pending after create- or update-content-campaign set its sources. A web page is read now (up to about 25 seconds) and comes back ready, or failed with the reason. A file comes back ready, failed, or still pending while Bolta extracts it; call again later to check. A brand-kit section is ready when the brand kit has text there. Only ready sources reach the writer. Changes nothing else on the campaign.

approve-many-posts

ChatGPT
Approve several posts the user said yes to, in one call (up to 25). Each item is exactly one single-post approval: same role check, same review rules, same stale-revision check, same scheduling. Every item must carry expected_fingerprint: the content_fingerprint of that post's row in list-inbox-items, so only the version the user saw is approved. Items are handled one by one and do not depend on each other: the result lists every post with ok, its new status and whether it is scheduled, or why it was not approved (stale_review, review_conflict, schedule_time_too_soon, not waiting for review, permission). If none was approved the call fails, with the same per-post list. Tell the user about every failure, and never retry a stale one without re-reading it and showing them the new text. Repeating the call is safe: posts already approved come back with already_decided=true. One post is one item: this is also the tool for approving a single post. A source=recurring row of list-inbox-items is approved by its review_id (the row's id) instead of post_id and expected_fingerprint; it schedules at its suggested time unless schedule_mode says otherwise.

approve-post

ChatGPT
Approve a post that is pending review. Requires admin or owner role. Supports scheduling modes: approve_only, use_suggested_time, or set_fixed_time.

approve-recurring-review

ChatGPT
Approve an AI-generated recurring post, optionally scheduling it at the suggested time.

archive-routine

ChatGPT
Retire a routine. Archiving is TERMINAL — an archived routine cannot be edited or re-armed, only duplicated. Its history (runs, approvals, published posts) is preserved; nothing is deleted. Confirm with the user before calling this; pause-routine is almost always what they meant.

bulk-create-posts

ChatGPT
Create MULTIPLE posts in one call. ALWAYS use this instead of calling create-post repeatedly whenever the user asks for more than one post (e.g. 'a week of content', 'make 7 posts') — looping create-post renders a separate card per post and the user only ever sees one of them. Use create-post only for a single post. Returns a task_id; the card polls bulk-create-status itself and lists every post's copy when the work lands, so you do NOT need to poll or restate the copy. Each post goes through the same approval, Safe Mode and autonomy rules as create-post — including not inventing the topic or destination: ask the user what the posts are about and which account(s) they target before calling this.

bulk-create-status

ChatGPT
Check the status of a bulk post creation task. The response's quality_blocked array is read live from the posts, so it stays accurate even for a post demoted after the task reported SUCCESS: each entry names the post, why it was blocked, and how to fix it. Repair those with update-post before submit-for-review.

confirm-plan

ChatGPT
Accept one stored proposal by proposal_id and materialize its already-previewed routine or campaign configuration. It cannot accept arbitrary operations or tool definitions. Save a plan the user said yes to, exactly as Bolta planned it (by its id). switch_on false (default): a routine is saved and off, a campaign is saved as a draft; nothing runs. switch_on true: saved, switched on and the first run or generation starts, which uses credits and leads to posts. Ask first, and ask separately before switch_on true. Confirming the same id again returns what was saved and runs nothing. A plan that still needs something (a connection, reply topics, keywords, the real material) is refused as plan_not_ready, naming what is missing.

create-content-campaign

ChatGPT
Start a themed, time-boxed body of content — "our spring launch", "Q3 thought-leadership push". Always lands in DRAFT: creating a campaign never arms it. Call set-content-campaign-status with action=activate when the user is ready, which is also what starts generation. A campaign is the BRIEF — what the content is about and when it runs. For one-off posts, use create-post or schedule-post.

create-engagement-opportunity

ChatGPT
Verify and persist a caller-supplied third-party target in the existing Hunter Mention pipeline. Use this when a target is known directly or when a platform, such as LinkedIn, does not provide public discovery. The operation is idempotent.

create-post

ChatGPT
Create ONE post (draft or scheduled). EXACTLY one — this tool is single-use per turn. If the request needs two or more posts — the same idea adapted for several platforms is the common case — call bulk-create-posts ONCE instead. Do not call this tool twice; do not loop it. Each call renders its own card and the user sees only the last one, so a second call does not add a post to the reply, it HIDES the first. That failure is invisible from here: both calls return 200 and the user quietly receives half the work. For a WordPress article use create-wordpress-blog-post instead, which handles the article settings, SEO metadata and Media Library upload in one call. NEVER invent the topic or the destination. If the user has not said what the post should be about, ask them before calling this — do not pick a subject for them. Likewise, if they have not said where it should go, ask which account(s), unless they explicitly asked for a draft only. A post created with invented content and no accounts is an orphan the user never asked for and cannot publish. If the workspace has Safe Mode enabled, non-Draft posts are automatically rerouted to Pending Approval. Images you generated in this conversation CAN be attached: pass them in the media parameter (no separate upload step needed) — or attach them to an existing post later via update-post's media parameter.

create-routine

ChatGPT
Create one saved routine configuration from a first-party template; this operation does not run the template or invoke a caller-selected tool. Save a repeatable outcome as a routine. This is the tool for "do this every weekday morning", "run this weekly", "keep doing this". It is the only automation doorway on this surface. Pass a template_key from list-routine-templates — that is how routines are made. The routine is always created as a DRAFT and never fires until activate-routine is called — creating one is not arming one. approval_policy may only ever TIGHTEN the workspace's policy: 'review_every_artifact' always forces human approval, while 'auto' is honoured only where the workspace already allows it. Ask the user whether replies/posts need approval and set it explicitly rather than assuming.

create-social-bucket

ChatGPT
Create a social bucket — a named group of connected accounts for one-shot multi-account publishing. Users phrase this naturally: 'create a bucket called Launch Team with my LinkedIn and X accounts' — resolve the accounts via list-accounts and pass their ids as account_ids.

create-voice-profile

ChatGPT
Create a new brand voice profile for a workspace. This is the first step in the voice bootstrap flow. Requires voice:write permission. The server completes the profile for you: it links the workspace's default Business DNA and copies its brand values / tone of voice / tone constraints / dos / donts into any field you left empty, defaults the content intent, assigns the workspace's unassigned connected social accounts to this voice, seeds starter platform rules and content structures for the platforms those accounts publish to, and derives the workspace's first Product from the Business DNA. Anything you pass explicitly is kept as-is. The response includes a 'bootstrapped' object listing what was filled in — report it to the user so they know what was set up. Either run extract-business-dna BEFORE this call, or just pass website_url and let this call do the scrape.

create-wordpress-blog-post

ChatGPT
Create a SERP-ready WordPress blog post in one call: resolves the WordPress account, creates the draft, sets WordPress article settings (title, excerpt, categories, tags, SEO meta), uploads featured/inline images to WordPress's real Media Library, and optionally submits for review, schedules, or publishes — all without separate list-accounts / update-post-platform-details / submit-for-review calls. Never invent the topic, images, or destination the user didn't ask for. If wordpress_account_id is omitted and the workspace has more than one WordPress account, this returns a typed selection_required result listing them — ask the user which one rather than guessing. 'SERP-ready' means technically and editorially prepared (headings, meta title/description, focus keyphrase placement) — never promise rankings.

cross-platform-analytics

ChatGPT
Get cross-platform performance summary for a workspace. Returns followers, engagement, and growth rate aggregated by platform. SEPARATE PIPELINE: these account-level numbers are synced from the platforms themselves and do NOT depend on Bolta's per-post publication records. So posts with has_publication_record: false in list-workspace-posts do not cause account-level zeros, and the two are not evidence about each other — a 0 here means this account genuinely gained no engagement in the window.

decide-suggestion

ChatGPT
Answer one of Bolta's suggestions (any row of list-suggestions) after the user has seen what it changes and said what to do. Ask first; never accept on your own. accept makes the change: a routine recommendation edits its routine as if the user did; a brand-wide suggestion makes every change it lists (it can change a voice, a routine's schedule, switch a routine on or off, or start a campaign; only the brand's owner or an admin may). decline turns it down and keeps the record. undo reverses an accepted one (a brand suggestion within 7 days). Bolta works out which kind the id is and which routine it is for.

delete-post

ChatGPT
Delete a post. Recoverable for 7 days — the post is hidden immediately and permanently destroyed after the undo window.

draft-engagement-reply

ChatGPT
Generate a Voice Context-aware reply for a persisted opportunity. CALL get-engagement-conversation FOR THIS OPPORTUNITY FIRST. Search returns a title and a link, not the discussion — drafting straight off a search hit produces a reply that is on-topic and answers nothing, which is the single most common way engagement reads as spam and gets an account throttled. Read the thread, then draft. The draft is stored as the existing Hunter Reply record and requires approval; nothing is posted here. Repeating the same opportunity returns the existing draft instead of generating again, so a retry costs nothing.

draft-inbound-reply

ChatGPT
Draft a reply to a comment or mention left on OUR OWN content. ALWAYS call this rather than composing the reply yourself. A reply you write directly is in YOUR voice; this one is in the workspace's, because it runs the Voice Profile and Business DNA the customer configured. Reading the comment with get-social-content or list-inbox-items is not a substitute — that is the input to this call, not a replacement for it. USE THIS, NOT draft-engagement-reply, for anything on our own posts. They route to different server-side writers and picking the wrong one is SILENT: draft-engagement-reply writes as an outsider joining someone else's thread, so used here it produces a reply that is on-brand, fluent, and reads as though a stranger wandered into our own comments. Nothing errors and nothing warns. draft-engagement-reply is for third-party conversations Bolta discovered; this one is for the people talking to us. Reads the comment AND the post it was left on — a terse comment ("How?", "price?") is unanswerable without the thing it is asking about — then writes in the workspace's voice. Stateless: creates NOTHING and sends NOTHING. It returns text with drafted_only=true. Show it to the user, and send the approved text with reply-to-social-content.

extract-business-dna

ChatGPT
Extract a Business DNA record (industry, audience, positioning, tone signals) by scraping a brand's website. Reaches the public internet and consumes AI credits (ai:generate). Creates a new DNA record on the workspace, and — when the workspace has no products yet — derives its first Product from the same scrape (returned as 'product_created').

generate-team-post-drafts

ChatGPT
Write the drafts of a team version the user has seen and said yes to: one draft per row, saved together as one group for review. Drafts only: nothing is approved, scheduled or posted, and Bolta charges for a draft only when it is later approved. Pass the id and the revision the user saw. Refused if the plan changed since, if a row is still unresolved, if the plan or a full review queue blocks writing, or if it was already started (a repeat writes nothing twice). It returns as soon as the drafts are queued, before any is written: call get-team-post-plan to see which drafts are saved, and never say they are written until it shows them.

get-account-analytics

ChatGPT
Daily time series for one account — followers and engagement over the last N days, one point per day, plus total followers change and growth rate. The 'show me my Instagram followers/engagement over 30 days' view. READING THE NUMBERS: quote totals.engagement for activity gained in the window and totals.engagement_lifetime for the all-time figure. NEVER add up the series to get a total — each point's likes/replies/reposts/quotes/engagement is that DAY's delta, and the *_cumulative fields beside them are running lifetime totals that would be double-counted many times over if summed. A null delta means no earlier snapshot exists to measure that day against (unknown, not zero); counter_reset: true marks a day where the lifetime counter fell, usually deleted posts or a reconnect.

get-business-dna

ChatGPT
Get a specific business DNA entry — the brand's identity record: colors, fonts, visual aesthetics, logo, tagline, values, and positioning. ALWAYS pull this before generating an image for a post and fold the colors/aesthetics/logo guidance into the image prompt so visuals stay on-brand.

get-content-campaign

ChatGPT
Read one content campaign in full: its brief (name, theme, audience, strategy, call to action, content mix, window), its status, and its derived results. Unlike the list, this read derives the campaign's performance, so prefer the list when the user is browsing and this when they ask how one campaign is doing. Read-only.

get-engagement-conversation

ChatGPT
Read a third-party target and useful thread context before drafting. Supports Reddit, Threads, Bluesky, Mastodon, and LinkedIn post URNs. Reddit accepts t3_ submission or t1_ comment fullnames; for bare IDs pass target_type.

get-engagement-status

ChatGPT
Get one engagement's stored status and canonical URL. When supported, verify=true checks that the externally published reply still exists.

get-my-capabilities

ChatGPT
Check what the current API key or agent is allowed to do in a workspace. permissions is the caller's role intersected with the scopes this key actually carries; denied_by says why each denied action is denied (role, workspace_plan, or key_scope).

get-post

ChatGPT
Get details of a specific post

get-post-platform-details

ChatGPT
Read the platform-specific settings attached to a post for one platform. Supports x, threads, reddit, linkedin, facebook, instagram, bluesky, mastodon, discord, tiktok, pinterest, youtube, wordpress — e.g. Reddit subreddit + flair, TikTok privacy + duet/stitch, YouTube title/privacy/tags, Pinterest board + title, Threads reply controls, WordPress title + SEO meta. The returned fields tell you exactly what update-post-platform-details accepts for that platform.

get-posting-rules

ChatGPT
Read the rules a post must meet BEFORE you write or save it, for this workspace's connected accounts or one platform: character limit and what happens to longer text, required media, fields you must set (Reddit subreddit and title, Pinterest board, YouTube video), TikTok privacy default, threads, carousels, markdown and link handling, scheduling. Call it once before drafting for a platform you have not written for in this conversation and write to the tightest limit. It only reads: it never saves, schedules or publishes anything.

get-routine

ChatGPT
Get one routine's full configuration: outcome, trigger and schedule, instructions, capabilities, approval policy, source bindings, recent runs and performance summary. Read-only. For routines whose output_config.artifact_type is reply or report, review_queue is always empty; pending drafts for reply routines are in list-engagements status=pending (web: Hub > Approvals).

get-social-content

ChatGPT
Read any public post or comment by URL and return it normalized: platform, author, text, canonical URL, target type, and surrounding conversation. The target does NOT need to have been discovered by a Bolta agent or Hunter campaign — this works on a link you found anywhere. Use it to inspect a target before replying, or to pull conversation context for drafting. READING IS NOT DRAFTING. Do not write the reply yourself from what this returns — a reply you compose directly is in YOUR voice, not the workspace's, and it silently bypasses the Voice Profile and Business DNA that are the entire reason the customer connected Bolta. Draft through a tool: draft-inbound-reply for a comment on OUR own content, draft-engagement-reply for a third-party opportunity Bolta discovered. Wired for reddit, threads, bluesky, mastodon, linkedin, and x.

get-team-post-plan

ChatGPT
Read a team version: its rows, and for each one whether its draft is not started, queued or being written, saved, held back (its account is not connected) or failed, with the reason when Bolta recorded one. Gives the requested and saved counts, the saved posts' ids and where to review them. A row counts as written only when its draft is saved. Changes nothing.

get-voice-context

ChatGPT
Use this before generating any content. Returns the compact, resolved voice context for a workspace (schema voice-context.v1): the selected or default Voice Profile merged with its linked Business DNA — business identity, writing rules, tone, persona, platform adaptations, knowledge availability — with provenance and missing-field indicators. This is the same brand identity the Bolta app and internal agents resolve, so content written from it matches the user's voice everywhere.

get-voice-profile

ChatGPT
Get details of a specific brand voice profile

get-workspace

ChatGPT
One workspace's details — including its timezone, which is the timezone to use when scheduling, plus its name, plan and settings. Read-only.

list-account-insights

ChatGPT
Per-account performance breakdown for every connected account in a workspace: followers, engagement, and growth over the window. Use this for 'which of my accounts is growing' — a finer cut than cross-platform-analytics, which aggregates by platform.

list-accounts

ChatGPT
List all connected social accounts in workspace. Renders a card with Socials and Buckets tabs — connected accounts with their health, plus the workspace's social buckets (named account groups).

list-buckets

ChatGPT
List the workspace's social buckets — named groups of connected accounts used to publish to several accounts at once (e.g. 'Launch Team' = LinkedIn + X). Each bucket includes its member accounts (id, platform, username). Renders in the accounts card's Buckets tab.

list-business-dna

ChatGPT
List Business DNA entries for a workspace. Each entry captures brand identity, target audience, and messaging pillars.

list-content-campaign-posts

ChatGPT
Every post belonging to a content campaign, oldest first. Membership is TWO edges unioned here: posts explicitly attached to the campaign, and posts its runs generated. That union is why this tool exists rather than filtering the post list — a generated post carries only the second edge, so a campaign can look empty from anywhere else. Read-only.

list-content-campaigns

ChatGPT
List this workspace's content campaigns — themed, time-boxed bodies of content, newest first. A campaign is a SUBJECT and a WINDOW (theme, audience, editorial strategy, call to action, content mix, start/end dates). Each row carries the campaign's status (draft | scheduled | active | paused | completed | archived) and how many posts belong to it, but not its results; call get-content-campaign for those. Read-only. This is the CONTENT campaign surface — it has nothing to do with LinkedIn ad campaigns or Hunter keyword monitors.

list-engagement-capabilities

ChatGPT
List the engagement actions each connected account can actually perform. Capabilities fail closed for invalid or reconnect-required credentials; mcp_ready identifies an end-to-end adapter rather than an internal-only primitive.

list-engagements

ChatGPT
List stored engagement Reply records with target, approval, publication, retry, error, and URL state.

list-inbox-items

ChatGPT
Everything waiting on the customer, in one call — the dashboard hub's full inbox. THIS IS ALSO WHERE COMMENTS AND MENTIONS ON OUR OWN CONTENT ARRIVE. When the user mentions a comment left on their post, a mention, or a reply they need to answer, look HERE: the row carries the platform, the account, the target id/url and the comment text — exactly what a drafting tool needs. Do NOT go hunting through list-workspace-posts for it; that lists OUR posts, not the comments people left on them. Rows are labeled by source: posts awaiting team review (source=team), recurring AI-generated posts (source=recurring), engagement reply drafts and mentions such as Threads (source=hunter), and reports (source=report). Use this when the user asks what's waiting for them, what was produced for them, or about replies/mentions to review. For ONLY the recurring queue use list-recurring-reviews. The total_team/total_recurring/total_hunter/total_report counts are always workspace-wide — they do NOT shrink when you pass source, so an empty filtered result can still show work pending in the other buckets. A row here is the INPUT to acting on it, not the act. In particular, a row carrying comment text is not permission to answer it yourself: a reply you compose directly is in YOUR voice and skips the Voice Profile and Business DNA entirely. To reply to a comment or mention on our own content, call draft-inbound-reply. To revise or send a reply Bolta already drafted (source=hunter), use update-outreach-reply / send-outreach-reply. To approve a post, approve-post.

list-recently-deleted

ChatGPT
Posts and agent jobs deleted in the last 7 days that can still be restored with restore-deleted. Each item carries its type (post|job), a label, when it was deleted, and how many days remain before it is permanently purged.

list-recurring-reviews

ChatGPT
List AI-generated recurring posts that are waiting for review.

list-reviews

ChatGPT
List all POSTS pending review in a workspace (team + recurring post reviews). Drafted outreach/engagement replies awaiting approval are NOT posts: list them with list-engagements status=pending (needs_approval=true) and approve/send them with publish-engagement-reply.

list-routine-runs

ChatGPT
Execution history for one routine: each run's derived status (running | awaiting_approval | scheduled | completed | partial | failed), what it produced, and any error. Run status is derived from the authoritative records on every read and is never a stored field. Read-only.

list-routine-templates

ChatGPT
Read the fixed first-party routine template catalog. Returns configuration data, not new tool definitions or executable operations. Selecting a template does not execute it. Use create-routine to save its configuration, update-routine to edit it, activate-routine to enable future triggers, and run-routine-now to start a run. Browse the first-party routine library. A RoutineTemplate is the USER-FACING behavior description: what outcome it produces, which triggers it supports, which sources it can consume, which approval policies are legal for it, and the default instructions it seeds. Start here when the user describes an outcome they want to happen repeatedly ("post a weekly recap", "reply to relevant Reddit threads every morning"), then pass the chosen key to create-routine as template_key. Read-only. Templates whose available_in_workspace is false cannot be created here yet — say so rather than attempting it.

list-routines

ChatGPT
List this workspace's routines — the automations it has configured. Each row carries the outcome, trigger, approval policy, status (draft | active | paused | archived), when it last ran and when it next runs. Archived routines are excluded unless asked for. This is the answer to "what is Bolta running automatically?" — routines are the only automation on this surface. Read-only.

list-scheduled-posts

ChatGPT
List all scheduled posts in a workspace

list-suggestions

ChatGPT
What Bolta suggests changing, from what people approved, edited and rejected. Two kinds, both returned unless kind narrows it. routine_recommendations (kind=routine): per routine, from how its drafts did (many rejected, heavily edited, piling up unreviewed, nothing published, ready to do more, idle); each row says what it would change (proposed_change), why, its routine_id and status (proposed, or applied and still undoable). bolta_suggestions (kind=brand): brand-wide, for example use only your own photos, avoid a repeated phrase, post less often, move a routine's time, add to a brief, start a seasonal campaign; each has a title, the reason and evidence, the changes it would make, its status and can_decide (only the brand's owner or an admin may accept). Use it after a review pass, or when the user asks why drafts keep missing. Changes nothing.

list-voice-profiles

ChatGPT
Retrieve all brand voice profiles for a workspace

list-workspace-members

ChatGPT
List the people on this brand's team: each person's id, name and role (owner, admin, creator, viewer). Client reviewers are not listed. Use it to turn a name the user gives into the id that create-routine and update-routine take in default_reviewer_ids. No email addresses. Changes nothing.

list-workspace-posts

ChatGPT
List posts in a workspace with optional filters, rendered as a posts card. For "how did my post(s) do" questions, call with status=published, a start_date/end_date window (e.g. today or the last 7 days), and include_metrics=true — each post then carries per-platform engagement (views, likes, comments, reposts). FRESHNESS: metrics sync nightly from the platforms; for posts published less than ~24h ago, tell the user metrics may not have synced yet — zeros on a fresh post mean 'not synced', NOT that the post performed badly or failed. UNATTRIBUTABLE POSTS: every post carries has_publication_record. When it is false the post has no linked account/platform, so no engagement can be attributed to it and its zeros are MISSING DATA, not poor performance (metrics_unavailable_reason spells this out). Exclude those posts when ranking top performers and say how many you set aside — never present their zeros as results, and never conclude from them that metrics are broken workspace-wide.

list-workspaces

ChatGPT
List all workspaces accessible by your API key

pause-routine

ChatGPT
Suspend a routine's trigger without losing its configuration or history. Reversible with activate-routine. Use this rather than archive when the user says 'stop for now' or 'hold off this week'.

plan

ChatGPT
Create a routine or campaign proposal, or recommend routine proposals. These operations are also exposed individually as plan-routine, plan-campaign and recommend-routines, each with its own input schema and annotations. Have Bolta plan the way its own chat does, reading the brand's accounts, voices, existing routines, best recent posts and connected sources. Nothing is created; each plan is a card with an id. Pass the user's words as ask: Bolta decides whether it is a routine (something that repeats) or a campaign (1 to 4 weeks about one thing), as its own chat does. kind only overrides that. kind=routine: one routine from what the user wants (ask may be empty or vague: status shaping returns a working plan, what was filled and from where, what is still open and at most two questions; settle them with revise-plan changes; suggested_switch names a better-fitting routine, offer it, never apply it unasked). kind=campaign: one campaign, 1 to 4 weeks about one thing (a launch, an event, a sale), with dates, posts per week, accounts, voice, angle, the pages it is built from and a week-by-week outline; ask is required, a result with a question means Bolta needs one answer first. kind=recommend: 3 to 5 routines (count), best fit first, each a different job, none repeating one the brand runs, each with why it fits, sized to review_load; skipped says what was left out and why; ask is the user's goals. Show the plan(s) and ask whether to save; save each with confirm-plan.

plan-campaign

ChatGPT
Have Bolta plan one campaign (1 to 4 weeks about one thing: a launch, an event, a sale) the way Bolta's own chat does: dates, posts per week, accounts, voice, angle, the pages it is built from and a week-by-week outline. Returns a plan card with an id; nothing is created. Pass the user's words as ask, their goals and audience as context, and any pages they gave as links. A result with a question means Bolta needs one answer first. Show the plan and ask whether to save it; save with confirm-plan.

plan-routine

ChatGPT
Have Bolta plan one routine from what the user wants, the way Bolta's own chat does: it reads the brand's accounts, voices, existing routines, best recent posts and connected sources, picks the schedule, accounts, voice and what each run makes, and returns a plan card with an id. Nothing is created. Pass the user's words as ask (empty or vague is fine: status shaping returns a working plan, what was filled and from where, what is still open and at most two questions; settle them with revise-plan changes) and anything else they told you as context. suggested_switch names a better-fitting routine; offer it, never apply it unasked. Show the plan and ask whether to save it; save with confirm-plan.

plan-team-posts

ChatGPT
Plan a team version: one idea written as one post per person or function, each in that person's voice, for the account that voice posts from. Returns the plan as rows (who, voice, account, angle) with an id and a revision. Saves the plan only: nothing is written, charged, scheduled or posted. Put the people and functions the user named in participants; Bolta never adds one. With no participants it uses the workspace's person voices and says so. A row Bolta could not place (a name that matches several voices or none, a voice with several accounts) comes back marked unresolved with the choices: ask the user, never choose for them. Show the plan to the user before generate-team-post-drafts.

publish-engagement-reply

ChatGPT
Explicitly approve and queue a stored reply through Bolta's existing execute_approved_reply task and platform adapter. Idempotent for an already-sent reply. Use get-engagement-status to wait for and verify the external result.

publish-post

ChatGPT
Immediately queue a post for publishing. Requires admin or owner role. Blocked when workspace Safe Mode is ON — use submit-for-review instead.

react-to-social-content

ChatGPT
Like, upvote, favourite, or boost ANY public post using a connected account. The target does NOT need to have been discovered by a Bolta agent — pass the public target_url you already have (or platform + native target_id). Supported today: reddit (upvote, downvote), bluesky (like, repost), mastodon (favorite, boost), x (like, repost). Threads and LinkedIn have no reaction API wired and fail closed with a typed capability error. Call list-engagement-capabilities for the live matrix.

recommend-routines

ChatGPT
Have Bolta recommend 3 to 5 routines for this brand, best fit first: each a different job (posts, articles, founder stories, customer wins, reusing top posts, replies, reports, routines fed by a connected source), none repeating a routine the brand already runs, each with one line on why it fits from the brand's own posts, routines and sources. Sized to review_load (drafts per week, waiting, decided lately): a full review queue gets fewer, smaller routines and why says so. Each comes back as a plan card with its own id; nothing is created. skipped says what was left out and why. Show them and ask which to save; save each with confirm-plan.

reject-post

ChatGPT
Turn down a post that is waiting for review, with the user's reason, so Bolta learns what not to write. Works for drafts a Bolta routine or campaign wrote and for posts sent for review (list-inbox-items source=team rows). The post is kept and marked Rejected: never deleted, scheduled or published. In a campaign it stays in the campaign and counts as rejected. Use the user's own words as the reason and add the categories that fit. If the user has not said why, ask them first and offer the reasons Bolta uses: Too salesy (too_salesy), Wrong tone (wrong_tone), Not engaging (not_engaging), Too formal (too_formal), Too casual (too_casual), Wrong format (wrong_format), Off-topic (off_topic), Says nothing (no_substance), Too long (too_long), Repetitive (repetitive), Other (other). Their answer is what Bolta learns from, and it cannot be added after the post is rejected. source=recurring rows work too: pass the row's post_id and the recurring review is turned down. Needs an admin or owner. Repeating the call is safe.

reject-recurring-review

ChatGPT
Reject an AI-generated recurring post and optionally provide feedback.

reply-to-social-content

ChatGPT
Reply to an existing third-party social post or comment using a connected Bolta account. THE TARGET DOES NOT NEED TO HAVE BEEN DISCOVERED BY A BOLTA AGENT FIRST — pass a public URL you found yourself (web search, the user pasting a link, another tool) and Bolta resolves it, publishes through the platform's native API, and verifies the result. No Hunter campaign, no opportunity record, and no pre-existing reply_id are required; do NOT use send-outreach-reply for this (that one is only for replies Bolta's Hunter already drafted). Works for a top-level reply to a post AND for a reply to a specific comment. Wired for reddit, threads, bluesky, mastodon, linkedin, and x. Returns status (published or queued), the platform reply ID, the canonical URL of the published reply, timestamp, and verification state. Replying twice to the same target returns the original reply with deduplicated=true instead of posting again.

restore-deleted

ChatGPT
Undo a delete: restore one soft-deleted post or agent job by type + id (find candidates with list-recently-deleted). Returns 404 if the item is live, already purged, or not in this workspace. Restoring a job does NOT bring back its Hunter/Engager campaign — the response says so when that applies.

revise-plan

ChatGPT
Revise one stored routine or campaign proposal, identified by proposal_id. This edits the proposal only; it does not create, activate or run a routine or campaign. Change a plan from plan before it is saved: change is a sentence ("Tuesdays instead"), changes sets exact fields agreed in the chat with no guessing; both apply the sentence first. Returns the new plan with a new id and what changed; the old id can no longer be saved. This supersedes the previous proposal and cannot be undone through this tool; confirm-plan must use the new id. No routine or campaign is created.

run-routine-now

ChatGPT
Start one run of an existing routine with its saved template and configuration. The server executes that bounded automation, which can generate content and, where the workspace permits, schedule posts or send replies. It does not accept a new operation or tool definition. Dispatch an active routine immediately, off-schedule. Identical execution to a scheduled run — same runtime, same approval policy, same idempotency — differing only in the recorded trigger. A second call within 30 seconds returns the same run with started=false. A call while a run is still going is refused with 409 workflow_already_running: say it is already running. Each run spends credits. Drafts and paused routines cannot be run; arm them first. Returns immediately with the run inline — run.id plus run.status, derived on every read — and started. Call list-routine-runs for that routine to see how the run progressed; there is no per-run polling tool.

save-post

ChatGPT
Write a new post or edit an existing one. NEVER public: the result is a Draft (or, for an edit, the post's current state that has not gone out). Without post_id it creates ONE draft; with post_id it edits that post in place (it keeps its id and history; omitted fields stay as they are). Scheduling and publishing are publish-post's job: save-post refuses scheduled_time, status and requested_action. Editing an Approved or Scheduled post sends it back to review, and the result says so (sent_back_for_review). A Published or publishing post cannot be edited. To take a Scheduled post off the schedule, send post_id with unschedule=true: it goes back to Draft with its content and settings, and the result gives was_scheduled_for (the time it no longer goes out at). To move it to another time, use publish-post with the new time. The result carries content_fingerprint, which approve-many-posts takes as expected_fingerprint. Several posts at once: bulk-create-posts. A WordPress article with its SEO settings: create-wordpress-blog-post. Never invent the topic or the destination: ask what the post is about and which accounts it goes to. Quality Blocked post: send the rewritten content with post_id and the check re-runs.

schedule-post

ChatGPT
Schedule a post for future publishing. Requires admin or owner role. If workspace Safe Mode is ON and the post is a Draft, it is rerouted to Pending Approval instead of being scheduled. AFTER SCHEDULING A BATCH: finish with one list-workspace-posts call (status=scheduled) so the posts card on screen reflects the new state. Cards are snapshots of the call that drew them — the bulk-create card still shows every post as it was at creation, so without that final call the user reads 'Draft' under a message saying 'Scheduled'.

search-engagement-opportunities

ChatGPT
Search for relevant third-party conversations and persist stable, deduplicated opportunities in Bolta's existing Hunter Mention pipeline. Public discovery is available for Reddit, Threads, Bluesky, and Mastodon. Mastodon discovery is hashtag-based (the query is folded into a single tag) because status full-text search is unavailable on most instances. Requires a Hunter/Monitor campaign already attached to the account.

send-outreach-reply

ChatGPT
Send a drafted outreach reply from the inbox to the person who was found (a public reply on their post or mention): approves the draft and queues it for sending. Pass content to override the draft text in the same call. reply_id comes from list-inbox-items rows with source=hunter (Bolta's outreach finder). Sending twice is prevented server-side — the send task only fires for pending approved replies and skips re-posting an already-replied mention. To edit the draft without sending, use update-outreach-reply.

set-brand-approval

ChatGPT
Change a brand's approval settings: autonomy_mode (Default automation level: manual, assisted, auto, aggressive; auto and aggressive let routines, campaigns and agent jobs publish without review unless something stricter holds them; an agent's own level can only be stricter) and safe_mode (Approve team posts: true holds every post for review). Owner or admin only. Read list-workspaces first and pass its approval.updated_at as expected_updated_at. Making review stricter applies at once. Loosening is refused if an owner turned this connection's approval switch off (connection_automation_permission=not_granted); otherwise call once without acknowledge to get the list of what would publish without review, show it to the user, and only after their yes call again with acknowledge=true. Client sign-off is never changed here (Bolta only), and posts made through this connection still need approval.

set-content-campaign-status

ChatGPT
Move a content campaign through its lifecycle. One tool, five actions, because the legal move depends on where the campaign is now: - activate — draft or scheduled -> active. THIS IS WHAT STARTS GENERATION for a campaign that has never run; it is not a bookkeeping flag. - pause — active or scheduled -> paused. - resume — paused -> active. A campaign that already produced content resumes as a pure lifecycle move and does not regenerate. - complete — active -> completed. It ran its course; reporting still reads it and nothing new attaches. Does not lead back. - archive — TERMINAL from any state. Non-destructive: the campaign and its posts are preserved, and restoring means cloning. Confirm with the user first; pause is almost always what they meant. An illegal transition is refused with the legal targets named, so read the campaign's current status first if unsure.

submit-for-review

ChatGPT
Submit one or more posts for team review. Changes post status to 'Pending Approval' and creates TeamPostReview records. Only 'Draft' and 'Needs Revision' posts can be submitted. A post rejected here for being 'Quality Blocked' is NOT a dead end — the per-post result carries quality_block.message with the specific problem; rewrite that post with update-post (which re-runs the check and returns it to Draft when it passes) and submit it again, rather than reporting the block to the user as final.

update-business-dna

ChatGPT
Partial update — omitted fields preserved (merge handled server-side). Use get-business-dna first to see current values. Edits the brand identity that drives image generation and voice context: tagline, colors, fonts, aesthetics, values, overview, local identity, story, image settings. Pass only the fields to change as a fields object.

update-content-campaign

ChatGPT
Edit a content campaign's brief — any of name, strategy_key, theme, audience, cta_text, cta_url, starts_on, ends_on, content_mix. Send only what changes. Status is NOT writable here: lifecycle moves are set-content-campaign-status, and results are derived, never set. Sending either is rejected by name rather than ignored.

update-outreach-reply

ChatGPT
Edit the text of a drafted outreach reply (source=hunter in list-inbox-items) before it is sent. Only unsent drafts can be edited (already-sent replies 409). Use send-outreach-reply to actually send it.

update-post

ChatGPT
Update an existing post's text, media, poll, schedule, or status. Omitted fields are left unchanged. Use this to attach an image generated in this conversation to a post that already exists — pass it in media. Updates happen in place: the post keeps its id and its full activity history, so never delete and recreate a post to change it. QUALITY BLOCKED RECOVERY: this is how you repair a post whose status is 'Quality Blocked'. Send the rewritten content and the quality check re-runs on it automatically — if it passes, the post returns to 'Draft' and the response carries quality_check.cleared=true; if it still fails, quality_check.message says exactly what to change. Never try to set status yourself to clear a block: pipeline-owned statuses are refused by design.

update-post-platform-details

ChatGPT
Set platform-specific settings on a post after creating it, for any of: x, threads, reddit, linkedin, facebook, instagram, bluesky, mastodon, discord, tiktok, pinterest, youtube, wordpress. Reddit: selected_subreddit_id, flair_id, flair_text. TikTok: privacy_level, disable_comments/duet/stitch. YouTube: video_title, privacy_status, tags, category_id, made_for_kids. Pinterest: board_id, title, description, link. Threads: reply_control, topic_tag. WordPress: title (the blog post's headline — always set it; otherwise the first line of content is used), meta_title, meta_description, excerpt, categories, tags, post_type, featured_media_id — set these before publishing a blog post. Pass the settings as a fields object; call get-post-platform-details first to see the exact fields a platform supports.

update-routine

ChatGPT
Patch one saved routine configuration; this operation does not execute the routine or change its lifecycle state. An active routine uses the new settings on future runs. Change a routine's configuration — schedule, instructions, approval policy, sources, outputs. Partial update: send only what changes. status is NOT writable here; use activate-routine / pause-routine / archive-routine so each lifecycle change is audited on its own. template_key and workflow_definition_key are immutable — create a new routine instead. An archived routine cannot be edited.

update-social-bucket

ChatGPT
Rename a social bucket and/or replace its member accounts. Users phrase this naturally: 'add my Instagram to the Launch Team bucket' — fetch the bucket's current members via list-buckets, adjust the list, and pass the full replacement as account_ids (ids from list-accounts).

update-team-post-plan

ChatGPT
Change a team version that has not been started: give a row another voice or account, remove a row, or add a row for a voice. Pass the revision the user last saw; if the plan changed since, the call is refused and the current plan comes back. Giving a row another voice makes it that person's row. Use it to settle an unresolved row with the user's answer. Saves the plan only: nothing is written, charged, scheduled or posted.

update-voice-profile

ChatGPT
Update an existing brand voice profile. Use this to refine voice rules, adjust tone, or evolve the profile based on feedback. Requires voice:write.

voice-generate

ChatGPT
Generate AI content using brand voice. Supports two modes: 1. Profile mode: provide voice_profile_id to use a saved voice profile. 2. Scratch mode: omit voice_profile_id and provide 'context' + 'niche' + 'business_name' to generate content inline from DNA/business context — no profile required. Ideal for outreach pipelines, demos, and cold prospecting. Consumes AI credits on every call.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/bolta"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/bolta)