Editor and CLI · stdio MCP · setup guide 03

Cursor
connects locally.

Set up Agent Decision Kit as a local Cursor MCP server, check its status, and run a bounded browser task with tool approvals enabled.

01 / Build once

Prepare the source checkout

npm install
npm run build

Keep the absolute checkout path ready. The published npm package is not available yet; when it ships, this setup will be shorter.

02 / Add the server

.cursor/mcp.json for a project or ~/.cursor/mcp.json globally

Restart Cursor after adding the server, then enable it in Agent's available tools.

Cursor MCP supports tools over stdio as well as hosted transports. The Cursor CLI uses the editor's MCP configuration and can list server tools with `agent mcp list-tools` after the server is approved with `agent mcp enable` when needed. This checks connectivity without a model-backed turn. This project uses local stdio for the browser, so Chromium runs beside the Cursor process instead of in a hosted service.

Configuration option

Source-checkout server entry

{
  "mcpServers": {
    "agent-decision-kit": {
      "type": "stdio",
      "command": "node",
      "args": [
        "/absolute/path/to/agent-decision-kit/dist/cli.js",
        "mcp"
      ]
    }
  }
}

03 / Verify

Check the connection

Open Cursor's MCP settings, confirm the server is connected, then inspect available tools in the Agent tool picker. If Cursor CLI is installed and the server is not approved, inspect its command and run `agent mcp enable agent-decision-kit`; then run `agent mcp list-tools agent-decision-kit`.

04 / Try a real workflow

Start with sample data

Fill the Release note draft field in the local demo. Do not submit it or click Delete draft.

Client setup reference: Cursor MCP documentation ↗

← Browse all agent guides

Cursorconnects locally.