ai.monnet/mcp

Approve a plan step on a motion. Requires editor role on the motion. Use get_motion first to see the plan and read the step's number.

Ask Monnet a question in the context of a specific motion. The question is sent as a private message (only you and Monnet see it). Monnet's response is returned directly.

Post a public comment on a motion, either top-level or as a reply within an existing comment thread (pass parent_id). The user is asked to approve the message before it is sent — nothing is posted without their explicit confirmation. Monnet may respond — if it does, the response is included in the return value.

Write a new playbook for a workspace: the instructions Monnet follows when a request lands on that recurring workflow. Call list_playbooks first. Show the draft to the person and get their agreement before calling this — it becomes how Monnet handles that workflow for the whole team. Owners and admins only.

Open a new thread in a Monnet workspace from a free-form prompt. A thread is a pre-motion conversation: Monnet picks it up in the background and shapes the idea with the author, and it becomes a motion — a doc, a plan, people to coordinate — only when it earns one. Returns the thread's id and URL immediately; call `get_motion` with the returned short_id to read where it got to.

Create a folder in the company brain. Only once a subject already carries several pages that want holding together — a brain the size of a team's is a handful of folders and a couple of dozen pages, and a taxonomy built ahead of the pages mirrors the sources instead of compressing them.

Create a page in the company brain. Call get_wiki_tree first: most material belongs on a page that already exists, and a brain with one page per source is a filing cabinet, not a brain. Links in the body are verified against the workspace before the page is written — a link that resolves to nothing rejects the whole call, naming the target. Give the page an icon: the tree and the sidebar draw one glyph per row, so a brain where every page looks alike can only be read by reading every label.

Delete a company-brain folder. Its pages are re-parented, not deleted — deleting a folder never throws away a document. The user is asked to confirm before it happens.

Delete a company-brain page, with its comments and its version history. Not reversible, and links pointing at it break. The user is asked to confirm before it happens, so do not ask them yourself first. Prefer editing a page that is wrong, or merging it into the one it duplicates — deleting is for a page that should never have existed.

Delete one of the company brain's raw source files, and the stored document with it. Not reversible. Every page that linked it is repaired in the same move — the link becomes [Source removed], and what the page says is left untouched. A source is the record a page was built from, so a page that turns out to be wrong is a reason to fix the page, not to remove its source. This is for something that should never have been uploaded. The user is asked to confirm before it happens.

Revise a playbook the workspace already has. Pass only the fields that change — the others are carried over. A field you do pass REPLACES its whole value, and no previous version is kept, so read_playbook first and revise what comes back. Owners and admins only.

Change a company-brain page: rename it, and/or replace an exact string in its body. Read the page first — the match is literal, whitespace included. When new material contradicts what a page says, correct the page at the source; never leave the stale fact and append a correction beside it. Links in the result are verified before the write lands, so a bad id rejects the call instead of leaving a dead link nobody sees. A page get_wiki_tree shows without an icon has none yet — give it one while you are here.

Fetch the user's inbox — the motions they follow, in two buckets: 'priority' is what is waiting on them (an open motion with a step assigned to them, not yet dismissed), 'active' is everything else they follow. Spans every workspace unless one is named. Use this when the user asks 'what's on my plate', 'what needs my attention', or similar.

Read a Monnet motion's full details — summary, body, status, priority, plan steps with assignees, member roles, threaded comments (each comment shows a short id usable as parent_id in the comment tool to reply in its thread), your own private 1:1 conversation with Monnet on it (never anybody else's), and the files attached to it, each with the id read_motion_file takes. Call this when the user references a specific motion by its short id or URL. It also reads a thread, whose whole content is that private conversation.

Read the rules this team's company brain is built on — how pages are typed and templated, when to create one versus extend one, and the link format. Read it BEFORE your first write to a workspace's wiki. It is the same text the nightly ingest job runs on, so a page you write matches the ones Monnet writes; a team that has rewritten it for their own brain is exactly the case where guessing goes wrong.

List every folder and page in a workspace's company brain, nested, with each page's id. This is the wiki's table of contents and the only place page ids come from — read it before deciding a subject is missing, because most material belongs on a page that already exists. Names and paths only: use read_wiki_page for a page's contents.

List motions in a workspace, filterable by status. Returns short ids so you can follow up with get_motion on a specific one. Default status is 'open'.

The playbooks a workspace has written for Monnet — a playbook is how the team wants one recurring workflow handled, and Monnet loads it when a request lands on it. Call this before writing one: a workflow that already has a playbook wants an edit, not a second copy.

List the raw material feeding a workspace's company brain: uploaded files, plus motions the team closed and shared workspace-wide. Shows what the next nightly run will read and what it has already digested. Link a source from a page's ## Sources section as [filename](?source=<id>).

List the Monnet workspaces you are a member of, plus any you have been invited to and not joined yet. Returns the workspace name and slug needed by other tools (get_motion, list_motions, create_thread, etc.). Call this first if you don't know the workspace slug — and before creating one, since an invitation means the team already has a workspace.

Take a source out of the company brain's queue, once what it says is in the wiki. Call it after folding a source into pages yourself — otherwise it stays queued, the nightly job digests it again in its own way, and writes over what you did. It is what closes the loop when you do the ingesting rather than leaving it to that job. Marking one already digested is not an error, and requeue_wiki_source undoes it.

Verify your Monnet MCP connection and list the API keys on your account. Useful for debugging setup issues.

Move a page or a folder into another folder, or to the wiki root. Links survive a move — they point at the id, not the path — so reorganising is safe.

Add a text document to a workspace's company brain as raw material — notes, a transcript, a spec. The nightly job reads it and folds what it says into the pages it belongs on. Push the source rather than writing the page yourself when the material is long or covers several subjects: compressing it across pages is what that job does. For a fact that clearly belongs on one page you have read, edit_wiki_page is more direct.

Read a file attached to a motion. Text, markdown, CSV, JSON and source files come back as their contents; Word, Excel, PowerPoint and PDF documents come back as their extracted text; images come back as the image itself. A PDF whose text cannot be extracted — a scan, or a protected file — comes back as a URL to open instead. Call get_motion first: it lists every attachment with the id and the filename this tool takes.

One playbook in full: its name, its description, and the body Monnet actually follows. Call this before edit_playbook — a revision written without reading the current body replaces it wholesale, and whatever you did not know was in it is gone.

Read one company-brain page's full contents, as markdown. Read the page before editing it — edit_wiki_page replaces an exact string, and the ## Sources section at the bottom has to be kept current.

Read one of the company brain's raw source files — the material a wiki page was compressed FROM. Use it to check what a page claims against what its source actually says, or to read something the nightly job has not digested yet. Text comes back in full; anything else, or anything long, comes back as a signed link to fetch. The first line always says which. Sources are never edited: correct the page, not the record it was built from.

Reject a plan step on a motion, with an optional reason. Requires editor role on the motion. Use get_motion first to see the plan and read the step's number.

Rename a company-brain folder. Safe — nothing links to a folder by name, and the pages inside keep their ids.

Put a digested source back in the company brain's queue, so the nightly job reads it again. Use it to undo a mark made too early, or to have a source re-read on purpose — after the team rewrites its schema, or when a page the source fed turns out to be wrong. It does not change any page: it only decides what the next run looks at.

Search INSIDE company-brain pages — their titles and their bodies. Use it to find where a fact already lives before writing it somewhere new; get_wiki_tree gives you every title, so reach for this when what you are looking for is in a page's contents rather than its name.

Update a motion's summary, body, priority, or plan. All fields are optional — only include the ones you want to change. For plan updates, provide the full list of steps (use get_motion first to see the current plan, then modify and send it back). A plan with groups cannot be written here: ask Monnet with ask_monnet instead.