Cursor

Cursor exposes an MCP config pane under Settings → MCP. Add the Plainwork entry there.

Add the server

Cursor → Settings → MCP. Click "+ Add new MCP server" and paste:

{
  "plainwork": {
    "command": "npx",
    "args": ["-y", "@plainwork/mcp"],
    "env": {
      "PLAINWORK_API_KEY": "pw_agent_…",
      "PLAINWORK_API_URL": "https://api.plainwork.xyz/v1"
    }
  }
}

Replace pw_agent_… with the token from Plainwork → Settings → Agents → New agent. See Authentication for the full walk-through.

Cursor saves the config and spawns the subprocess. You may need to toggle the entry off and on once for Cursor to pick up the new env.

Verify

Open a Cursor chat and ask:

What Plainwork collections are available?

The assistant invokes list_collections. Cursor shows the tool call in the chat transcript.

Pin a version

{
  "plainwork": {
    "command": "npx",
    "args": ["-y", "@plainwork/mcp@0.1.0"],
    "env": { "PLAINWORK_API_KEY": "pw_agent_…" }
  }
}

Trust and auto-approve

By default, Cursor's user approves every tool call in the UI and every write in Plainwork goes through the workspace approval gate. That's two gates — acceptable while getting comfortable, annoying in steady state. When ready:

1. Flip "Auto-run" on the Plainwork server in Cursor's MCP settings.
2. Flip "Auto-approve" on the agent in Plainwork → Settings → Agents.

Both must be on for writes to land immediately. Leave both off if you want the full human-in-the-loop experience.