AI assistants (MCP)

AI assistants join your documents through MCP (Model Context Protocol) — the standard way tools like Claude, Cursor, and VS Code connect to outside services. A connected assistant can list and read your documents, search them, answer questions left in comments, reply to comment threads, propose suggestions, and update data files — always as itself, with its own name on everything it does.

Connecting an assistant

Open ⚙ Settings → Connect an AI assistant…. You get the same list of assistants either way you’re using DraftMesh, but what the panel can do for you differs:

• In your browser, on DraftMesh cloud — the panel shows the Cloud setup line for each assistant: Claude, Claude Code, Cursor, VS Code, Codex and Windsurf. Copy the one for yours into its MCP settings and sign in when it asks. DraftMesh can’t see your computer from there, so it says nothing about what you have installed and writes nothing for you.
• Running DraftMesh on your own machine — the panel additionally detects which assistants are installed and offers each one both ways to connect.

The two connections:

• Cloud — connects the assistant to your synced documents, from any device. It signs in with your account, nothing is installed, and its actions are recorded as acting on your behalf. Requires being signed in.
• Local — connects the assistant to the documents on this machine. No account needed; DraftMesh must be running. Click Connect to have DraftMesh write the configuration for you, or Copy the shown snippet and add it to the assistant’s own settings yourself. The snippet is displayed in full — what you see is exactly what gets written.

After connecting, restart the assistant so it picks up the new connection.

On your own machine each assistant’s card says Installed on this machine or Not detected, and whether DraftMesh can see a connection for it — for some assistants it honestly can’t, and says so rather than guessing. A card DraftMesh set up itself offers Reconnect and Disconnect; Disconnect removes only the entry DraftMesh wrote.

What agents can — and can’t — do

Agents get a deliberately bounded set of verbs:

• Can: read documents and history, search, add comments and replies, answer questions, resolve or reopen markers, propose suggestions, ask a named person to sign off on a document, save documents where allowed (data files, or documents they’ve been granted edit on), and — where the connection offers it — create a new, empty workspace.
• Can’t, ever: accept or reject suggestions, approve documents, restore old versions, or manage sharing. Deciding is a human act — an agent proposes, a person accepts. This boundary is enforced by DraftMesh itself, not by the assistant’s good manners.

Asking for sign-off sits on the proposing side of that line: an assistant can record the ask and put it in your queue, but only you can answer it.

On shared cloud workspaces, a registered agent’s default capability is suggest — it can read, comment, and propose, but its text only enters a document when a human accepts it. Granting an agent full edit rights is an explicit, per-document decision by an admin, with its own confirmation step.

Asking an agent to look at a document

Everything above is the assistant coming to DraftMesh. The reverse — you asking an agent to go and work on the document in front of you — lives at the bottom of the Comments panel, under Agent activity. It is there only where there are agents to ask: on a DraftMesh with no registered agents and no request history it renders nothing at all rather than an empty box.

Request review from an agent opens a short form:

• Agent — which one. If it can’t be reached yet, the form says so — “This agent has no delivery endpoint, so it can’t receive requests yet.” — rather than accepting a request that goes nowhere.
• Scope — Whole document, or A section and then which heading.
• Instructions — what should the agent look at? This is the whole brief; there is no other channel.
• Send request, and the request appears in the list below as it progresses.

If the agent doesn’t have access to the document yet, sending grants it Suggest — enough to read, comment and propose, and never more. Giving an agent Edit stays a deliberate act with its own confirmation in the Share panel. Where you aren’t allowed to grant, the form says so instead of failing at send time.

Each request shows what became of it: Pending, Delivered, Completed, Failed, or Expired. A completed review offers Show the comment this answers, so the result is one click from the thread it belongs to. Cancel request withdraws one that hasn’t been picked up, and a request replaced by a newer one is marked “Superseded by a newer request.” rather than left looking stuck.

Connect an agent…, beside it, is the other question — not “do this now” but “do this whenever”: a standing rule that wakes an agent every time the document changes. That is administration rather than day-to-day work, and it is admin-only; see Admin: Governing AI agents. Rules connected to this document are listed here as Connected agents, and every time one fires it appears in the same list as a request.

Attribution and audit

Everything an agent does is labeled as agent work — in comment threads, in document attribution, and in version history. You can always tell whether a change came from a person or an assistant, and which one.

Putting agents to work

• Assign a comment to a connected agent to hand it a task.
• Leave a question in a comment and let the agent answer it in-thread.
• Have an agent keep a data file fresh — metrics, statuses, inventories — while humans work in the prose (see Data-driven pages).
• Or skip the connection entirely: ✦ Form reply exports a document’s comments as a paste-ready handoff for any chat assistant.