ros2-mcp

ROS2 ↔ MCP bridge: let any AI agent (Claude Code, OpenClaw, Cursor) read ROS2 topics, call services, and wrap telemetry into verifiable proof bundles.

Two modes:

• Mock mode (default) — simulated robot, zero ROS2 required. Try it now.
• ROS2 mode — real topics/services via a Python sidecar (rclpy).

Quick start (mock, 2 minutes)

npm install
npm run build
npm start

Register with your agent:

claude mcp add ros2-mcp -- npx tsx src/index.ts

Then ask:

"List the robot topics, then tell me the current GPS position." "Move the robot forward, then wrap the telemetry into a proof bundle for task task_ros2_001."

Tools

Tool	Purpose
list_topics	list ROS2 topics
read_topic	latest value of a topic
call_service	call a service (e.g. cmd_vel)
robot_status	status + GPS + telemetry in one call
wrap_proof	wrap telemetry into an ai2robot proof bundle draft (hash-chained)
chain_fingerprint	genesis hash for a task

Real ROS2

Requires a ROS2 environment (tested pattern: ros:humble docker image):

docker run -it --rm ros:humble
apt-get install -y python3-pip && pip install rclpy  # usually preinstalled

Then run the server with the sidecar:

AI2ROBOT_ROS2=1 npm start

The sidecar (python/ros2_bridge.py) subscribes to all topics generically and exposes them over stdio JSON-lines. It is a reference implementation; bring your own transport if you need typed QoS or specific vendors.

Why proof mode

wrap_proof emits bundles in the ai2robot proof format (hash-chained action log + GPS traces; Ed25519 signature added by the executor's key). Any fleet using this bridge can become an ai2robot executor without changing hardware — the bridge is the adapter kernel.

License

MIT