Roo Code

Roo Code has no hook system, so Comity cannot see or block its edits as they happen. comity init connects Roo to the Comity MCP server and adds rules that tell the agent to declare its work and check for conflicts through Comity's tools.

Set up

cd your-repo
comity init --tools roo
git add .roo/mcp.json .roo/rules/comity.md
git commit -m "Set up Comity for Roo Code"

Reload the VS Code window and make sure the comity server is enabled in Roo's MCP panel. There is no token in the repo: comity login is enough.

What init writes

.roo/mcp.json

{
  "mcpServers": {
    "comity": {
      "command": "node",
      "args": ["-e", "<launcher>", "mcp"]
    }
  }
}

Roo starts comity mcp, a local proxy to the Comity server that authenticates with your comity login (or COMITY_TOKEN). The entry is the same on every OS, so the committed file never changes between teammates; it needs Node on PATH (the CLI needs it anyway). <launcher> stands for a one-line Node script that starts comity mcp through your shell, so it finds the npm comity shim on every OS. Other servers in the file are kept.

.roo/rules/comity.md

The coordination rules inside a <!-- comity:start --> … <!-- comity:end --> block. Roo adds every file in .roo/rules/ to its system prompt. Re-running init updates only that block.

Verify

comity doctor

Expect roo mcp comity server in .roo/mcp.json and roo rules .roo/rules/comity.md is current. Doctor warns about an entry from an older init that reads ${env:COMITY_TOKEN} while it is unset.

Limits

• No hooks: Comity only knows what the agent reports through MCP tools (intent_declare, scope_check, …). Edits are never blocked automatically.