Zed

Zed's built-in agent uses Comity through MCP and its project instructions. Zed doesn't document agent hooks, so nothing is automatic: no briefings, no recorded edits and no blocks. The agent only knows about Comity because the instructions tell it to call the tools. See What works without hooks.

Set up

cd your-repo
comity init --tools zed
git add AGENTS.md
git commit -m "Set up Comity for Zed"

Then add the MCP server to your user settings (zed: open settings file). It runs the stdio bridge, so there's no token in any config: comity login is enough.

{
  "context_servers": {
    "comity": {
      "command": "node",
      "args": ["-e", "const c=require('child_process').spawn('comity '+process.argv.slice(1).join(' '),{stdio:'inherit',shell:true});c.on('error',()=>process.exit(1));c.on('exit',(k)=>process.exit(k===null?1:k))", "mcp"]
    }
  }
}

comity init prints this entry. The node -e launcher runs comity mcp through your shell and works the same on every OS; Node must be on your PATH, which the CLI needs anyway. Zed documents context_servers for the user settings file only, so init doesn't write it into the repo.

What init writes

Comity's coordination instructions go into a block marked <!-- comity:start --> … <!-- comity:end --> in the project instructions file. Zed reads only the first of .rules, .cursorrules, .windsurfrules, .clinerules, .github/copilot-instructions.md, AGENT.md, AGENTS.md, CLAUDE.md and GEMINI.md it finds (Zed instructions). init writes into that file, or creates AGENTS.md if the repo has none. The rest of the file stays as it is.

Verify

comity doctor

Expect zed instructions Comity block in … and zed mcp context_servers.comity in …/zed/settings.json. In Zed, Settings → AI → MCP Servers should show comity with a green dot.

Limits

• No hooks: no automatic briefings, conflict warnings or blocks, and the agent's edits aren't recorded.
• Tools can't infer the repo. The instructions tell the agent to pass repo (owner/name).
• Zed asks before every MCP tool call unless you allow them in agent.tool_permissions (for example, mcp:comity:team_context).