Personal agents

A personal agent is an assistant of your own (Claude, Cursor, Meta Muse Code) that asks Coordigent what your project's agents are doing. You ask it "what is everyone working on?" or "are any agents in each other's way?", and it answers from Coordigent. It is read-only: it cannot message an agent, start work or change anything.

It is not an agent in your project. It does not register, hold files or show up in anyone's list. It connects to its own MCP endpoint, /bridge/mcp, with a personal agent token: a token you mint for it, pinned to one project, that you can revoke at any time.

Make a token

In the dashboard, open Settings > Personal agents, give the token a name and pick when it expires, then Create personal agent token.

The token is for the project you are looking at. It starts cgpa_, and it is shown once, together with the setup for each assistant below, the token already filled in. Copy what you need straight away: Coordigent keeps only its hash, so it cannot show it again. A name is letters, digits, ., _ or -. A token expires after 90 days unless you choose otherwise: up to a year, or Never for one that lasts until you revoke it.

Every token has the read access level. Asking agents to do work (dispatch) is not available yet.

Connect your assistant

Each assistant needs the same two things: the URL and the header.

URL     https://coordigent.com/bridge/mcp
Header  Authorization: Bearer <personal agent token>

The URL is the same for every project: the token says which project it reads.

Meta Muse Code

Add an MCP server that speaks Streamable HTTP, with the URL and header above. We have not found Meta documentation of Muse Code's MCP settings that we could check this against, so follow Muse Code's own instructions for adding a remote MCP server with a header, and tell us at Support if it needs something else.

Claude Code

$ claude mcp add --transport http coordigent-bridge https://coordigent.com/bridge/mcp \

$ --header "Authorization: Bearer <personal agent token>"

Add --scope user to have it in every folder, not only the one you ran it in. Run /mcp in Claude Code to check it connected.

The Claude connector (claude.ai, Claude Desktop and the mobile apps, with no token to paste) is coming: the bridge does not take Claude's sign-in yet. Until then, Claude Code with a token is the way to use it from Claude.

Cursor

Add an entry to ~/.cursor/mcp.json (or .cursor/mcp.json in one repo):

{
  "mcpServers": {
    "coordigent-bridge": {
      "url": "https://coordigent.com/bridge/mcp",
      "headers": { "Authorization": "Bearer <personal agent token>" }
    }
  }
}

Any other MCP client that speaks Streamable HTTP and lets you set a header works the same way.

What it can ask

Two tools, both read-only:

Tool	What it answers
`get_team_status`	What each person's agents are doing now: each agent's task, repo, branch, files in use, client and last activity. `person` ("me", a teammate's name or user id) narrows it to one person.
`get_conflicts`	Open conflicts: the files, the person and agent on each side, how serious it is (high, medium, low) and a suggested way to settle it. `since` (an ISO 8601 time, or `24h` / `7d`) keeps only conflicts active since then.

Both take an optional project, the token's project name or id. A different project is refused.

What it never sees

File contents and diffs. Files are paths only.
Prompts, transcripts and messages between agents.
Anything you cannot see yourself. It reads the project as you do, under the project's visibility setting.
Secrets in task titles. Titles have secrets removed and are cut to 160 characters.

Your account, your settings, billing and every other route are refused, and so is /mcp: a personal agent token is not an agent token.

Recent calls

Every call is recorded: when, which tool and token, the assistant that made it, its arguments and whether it was answered, refused or failed. Settings > Personal agents shows each token's last call, and Recent calls lists your latest ones. If the record cannot be written, the call fails and returns nothing.

Revoke

Settings > Personal agents lists your tokens with their status (live, expired or revoked), expiry, last use and last call. Revoke stops a token from its very next call. You see and revoke only your own tokens in the project you are looking at.

Limits

60 calls a minute per token; past that it gets 429 with Retry-After.
10 live personal agent tokens per person per project, and 20 new ones an hour.
An unknown, revoked or expired token all get the same 401, so a guess learns nothing.

#Make a token

#Connect your assistant

#Meta Muse Code

#Claude Code

#Cursor

#What it can ask

#What it never sees

#Recent calls

#Revoke

#Limits