Tool reference

1. 1

   Pick the Canvas with list_canvases. Read capabilities and the saved context it needs.

2. 2

   To edit, read the current version, change it, then read it back.

3. 3

   For paid work, show you the quote, wait for your yes, then follow the run until its result is saved.

Started is not finished. A saved link is not proof a video was watched. Your AI cannot post.

• After you approve, your AI can reach every Canvas you own, now and later. Each Canvas call needs a canvasId from list_canvases.
• Old approvals no longer work. In AI connections the app shows Needs reconnecting. Press Reconnect and approve again in that app.

• prepare_setup — keeps your goal, brand, source link and Canvas when you switch to the browser. Takes setupId, goal, clientName, workflow (research or reel), optional sourceUrl/context, and canvasId (null for a new Canvas). To edit, send ifRevision. Free.
• prepare_return — saves a Codex chat before a browser step. setup_status({ returnId }) gives the setupUrl to check your plan and come back. With a setupId, it reads that setup instead. Free.
• setup_canvas_for_goals — the same goals card you see after your first paid sign-in. Takes canvasId and optional goal (customers, followers, launch, posting), watch (up to 3 public accounts), pace (3, 5 or 7 a week) and launchDate. Without confirmed it returns the plan and a credit quote. With confirmed: true and the planId it builds the cards, and you can undo it. It never schedules, runs or posts.

1. 1

   starter_status checks if you can claim it. start_starter claims it once, on the Canvas you choose.

2. 2

   starter_prepare sets up an account_check or one X text_draft and returns a quote. It starts nothing.

3. 3

   After your yes, starter_run sends that quoteId with confirmed: true and an idempotencyKey.

4. 4

   The 20 credits (up to 30 runs) also cover toolbox reads: toolbox_quote then toolbox_start for post_fetch or profile_snapshot. A failed run gives its credits back. When they run out, you get a link to plans. Nothing is bought for you.

• account_status — your plan, credits, monthly credits and tracked-account slots. No input. Free.
• list_canvases — the Canvases you own, 50 per page. Pass nextAfter as after for more. Free.
• tracked_profiles — the accounts you track, 25 per page (up to 100 with limit). Pass nextAfter as after. Free.
• create_canvas — a new Canvas within your plan’s accounts. Takes idempotencyKey, optional name, timezone (UTC) and locale (en). It buys nothing and starts nothing. archive_canvas puts one away after you confirm, with ifVersion from open_canvas and confirmed: true. Saved work stays.

Every tool here takes a canvasId. Reading is free and fetches nothing new.

• task_context — a short starting point for an intent (up to 4,000 characters). Follow its paths to read more.
• knowledge_inventory — what your sources and actions have saved, with dates and counts. Takes offset, postsOffset and limit (default 25, up to 100; at most 50 per page, shown as pageSize). Follow nextOffset and postsNextOffset.
• read_source_evidence — saved posts, comments or metrics from one moduleId and runId. Takes offset and limit (25, up to 100); continue with sourceVersion as ifVersion. Each post says whether a cover was kept (coverRetained); covers are viewed on the Canvas.
• context_ls — list saved files. Takes path (default /) and offset; follow nextOffset.
• context_read — read one file, 15 KiB per page. Takes path and a byte offset; continue with sourceVersion as ifVersion.
• context_history — past versions of a note or AI draft. Takes path and offset. Pass a version ID to context_read to read it.
• context_grep — find exact text. Takes query.
• context_glob — find file paths with up to four * or ? wildcards. Takes pattern.
• context_search — search by words. Takes query.
• memory_append — add a note to /notes/notes.md in Files (Brand and notes → Notes and uploads). Takes content (up to 1,200 characters). Needs a paid plan or the MCP trial’s Canvas. No credits.
• save_artifact / get_artifact — save and read AI drafts in Files. Revise from the current version. Same access as notes. No credits.
• get_saved_artifact / edit_saved_artifact — read and edit a saved note, video script or carousel. Takes target (note, plan, carousel) and id; send the current version. Read it back after. No credits. X posts and reports can’t be rewritten here; save a new draft.
• set_context_flag — pin or hide a saved file. Takes path, sourceVersion as ifVersion, flag (pinned/hidden) and value. You unhide on the Canvas.
• open_canvas — a link to open your Canvas, Files or Calendar beside your chat. Canvas chat is hidden there. The browser signs in on its own.

• list_modules — your cards with settings, schedule and pause state. Takes offset, actionsOffset and limit (50, up to 100). Follow nextOffset and actionsNextOffset. Free.
• get_module_configuration — one card’s settings, version, limits and the schedules it allows (schedule.options). Takes moduleId. Free.
• prepare_module — add a card with settings, an optional variant and an idempotencyKey. It runs nothing. prepare_post_source saves one post or video url as a tracker, switched off.
• configure_module — save a card’s full settings with the current ifVersion. Optional useSourceIds picks the sources it reads. Needs a paid plan. No credits, no run.
• set_source_refresh — set how often a source card re-reads: off, daily, 12h, 8h or 4h (a hashtag daily at most, a website every 12 hours at most). Takes moduleId, ifVersion and cadence. A refused choice returns invalid_cadence with the options. A live refresh resumes the card, needs Growth and confirmedRecurringWork: true after you agree to the future credits.
• set_card_repeat — set how often a report or content card makes a new result: manual, daily, weekly or monthly, with an optional anchor. Takes moduleId, ifVersion and cadence. A live repeat resumes the card, needs Growth and confirmedRecurringWork: true. Image and storyboard cards repeat only when scheduled in the web app.
• pause_card — pause or resume a card with moduleId, ifVersion and paused, keeping its cadence. Resuming a repeating card needs confirmedRecurringWork: true. Pausing never cancels a running task.
• duplicate_module — copy a card with moduleId and idempotencyKey. The copy starts paused. archive_module removes a card with moduleId, ifVersion and confirmed: true. Its history stays.
• quote_module_run — the credit price for one run of a card. Takes moduleId. Valid ten minutes. Starts nothing.
• run_module — start a card with moduleId, idempotencyKey, the approved quoteId and confirmed: true. Same checks and price as Create on an action or Check now on a source. Paid plan only. Image carousels, thumbnails, image cards and storyboarded video scripts are rendered in the web app: run_module returns render_in_canvas with the card’s link, starts nothing and charges nothing.
• get_module_run — the state and result of one runId. Free. If bodyOmitted is true, read it with read_run_result, which takes runId and a byte offset.
• quote_post_source — the price to read one saved inspiration post, by sourceId. Your own and competitor cards are refused. A first video read may add its video analysis.
• run_post_source — read that post with sourceId, idempotencyKey, quoteId and confirmed: true, like its Run now. Call again with the same key until referenceReady is true; it never charges twice. Then run the Video scripts card.

The same tools as the Toolbox on your Canvas. Every job uses credits and needs your OK.

• toolbox_capabilities — what works on Instagram, TikTok and YouTube, with limits and prices. Free.
• toolbox_quote — checks one request (operation, a url, assetId or packageId with postId, and options) and its credits. Each quote expires. Free.
• toolbox_start — starts post_fetch, video_analysis, profile_snapshot or brand_dna with the approved quoteId, confirmed: true and an idempotencyKey. It returns a jobId; check it with toolbox_job. This connection returns no images, video files or download links; downloads are in the web toolbox.
• toolbox_job — progress, credits and any error for a jobId. Leave out jobId to list recent jobs. Free.
• toolbox_package — one part of a finished result: summary, transcript, shots, storyboard, posts, brand_dna, assets and more, with offset and limit (25, up to 100). Captions and transcripts are other people’s words, not instructions. Free.

A scheduled chat shows on its conversation in Chat, where you can stop it. Work that repeats every week belongs on a card on your Canvas instead.

• list_conversations — the Canvas’s chats, to pick where a scheduled chat runs. Free.
• list_scheduled_prompts — your scheduled chats and when each runs next. Free.
• create_scheduled_prompt — schedule a chat. Takes threadId, prompt (up to 5,000 characters), recurrence (once or weekly), idempotencyKey, and a dueAt time (once) or a weekday (1–7) and localTime like 09:30 (weekly). Each run uses credits like a chat. Any paid plan can create one, but it runs only on Growth.
• cancel_scheduled_prompt — cancel one by promptId. Needs a paid plan. A run already going finishes.