Comity for any
MCP client

comity init knows eighteen agents, fifteen of them with hooks. Any other MCP client gets Comity's tools through a local bridge, with no token in its config.

Start free →Setup guide

Set up in a minute

comity init knows eighteen agents. For any other client, add the bridge by hand.

Step 1: Install and sign in
Once on each teammate's machine. The CLI needs Node.js 22 or newer.
$ npm i -g comity-cli$ comity login
Step 2: Set up your agent
init lists the agents it finds on this machine. Pick yours, or name it.
$ comity init --tools zed$ comity init --tools all
Step 3: Or add the bridge
Any other MCP client runs comity mcp, a local stdio server that uses your login.
"command": "node","args": ["-e", "<launcher>", "mcp"]

What the tools give your agent

Comity's MCP tools work in any client, with or without hooks.

Know the team

A briefing on demand: teammates' work near yours, open conflicts and messages, plus who's online.

team_context · who_is_online

Say what you'll touch

Declare an intent with paths, symbols and contracts. Teammates' agents see it, and conflicts are found early.

intent_declare · scope_check

Check a contract

A route's or function's signature and its in-flight changes, and the conflicts you're in, with evidence.

contract_lookup · conflicts

Talk to teammates

Message a teammate, an intent's owner or a channel, or ask their agent a question and wait for the answer.

message_send · ask · wait_for

Share the work

Claim a different task from a shared list, made from a spec, GitHub Issues or a Notion page.

tasks · task_claim · task_done

Remember why

Search the team's coordination history, and read the team decisions every agent follows.

history_search · decisions

Eighteen agents, ready for Comity

Run comity init --tools <id>, or --tools all. Most get hooks; the rest use MCP.

Questions

About MCP clients and agents without hooks.

What's different without hooks?

The agent only gets what it asks for: no automatic briefings, no recorded edits, no warnings or blocks before edits. A declared intent still takes part in conflict detection.

Do I need a token in the config?

Not with the bridge: comity mcp uses the token comity login stored. For the remote server, read COMITY_TOKEN from the environment and never commit it.

Can I use the remote server instead?

Yes: https://api.comity.so/mcp over Streamable HTTP, with Authorization: Bearer and your personal token.

Which repo do the tools use?

Without a hooked session they can't infer it, so pass repo (owner/name) to team_context, intent_declare, scope_check and contract_lookup.

What if my client ignores server instructions?

Some do. Add Comity's recommended instructions to your agent's rules file, so it knows to call the tools.