Skip to main content
The ai-cmo.dev MCP server (ai-cmo-mcp) is a local stdio process that gives an AI assistant (Claude Code, or any MCP client) read/write access to an ai-cmo.dev workspace by proxying the REST API.

​
Architecture

The server runs on your machine. It never exposes an HTTP endpoint — communication is exclusively over the MCP stdio transport. The binary connects to the ai-cmo.dev web app over HTTPS (or a local dev instance) and authenticates via an API key or a device authorization flow.

​
MCP vs REST API directly

​
Capabilities

  • 10 tools — reading scores, scoreboards, recommendations, runs, reports, workspace identity, and assistant chat; one mutating tool for guarded recommendation lifecycle transitions.
  • No resources or prompts — the server only exposes tools. Every interaction is a tool call.

​
Requirements

  • Node.js 20 or later (for npx and the ai-cmo-mcp package)
  • The ai-cmo.dev web app running at http://127.0.0.1:3000 (or a remote instance via AICMO_BASE_URL)

​
Next steps

Install the server

Add the server to your MCP client and authorize it in the browser.

Authorization

How credentials resolve and how the device flow works.

Tool reference

All 10 tools, their parameters, scopes, and example results.

Usage examples

Natural-language prompts and a worked multi-step workflow.