Skip to main content

Any robot. Any AI. A single connector.

You already know what the robot should do. Now you can just say it.

Get Started in 5 MinutesStar on GitHub
pipx install git+https://github.com/robotmcp/robotmcp_server.git
natural language

Claude

Move 4 meters forward, then shake a hand.

A prompt typed to Claude becomes ROS topic calls, which drive a drone, a quadruped, a mobile manipulator, and a warehouse unloading arm.

1,200+ GitHub starsOfficially in the MCP RegistryBuilt with KUKA · Doosan · Intrinsic (Google) · Linux Foundation

One interface. Four kinds of robot.

Every clip below is the same MCP server, driven by a different model, commanding a different machine. No per-robot integration work.

  1. Aerialvia Gemini

    Fly in a circle and land at the takeoff point.

    One line of instruction, and the flight path plus the landing back at the takeoff point are worked out for you.

  2. Leggedvia Cursor

    Move 4 meters forward, do various flips and shake a hand.

    Three behaviours in one sentence, carried out in order: travel, then flips, then a handshake.

  3. Mobile manipulationvia Gemini

    Grab a Coke from the refrigerator and go to the living room.

    Driving and grasping in a single request, across two rooms, with no per-step scripting.

  4. Industrialvia Claude

    Test the gripper on the unloading robot.

    The same interface reaches a warehouse unloading arm, driven from an ordinary chat window.

Built on an open protocol

  • Talk to your robot

    Command, debug, and monitor robots in natural language. Instructions become ROS topics, services, and actions automatically.

  • Any LLM, any robot

    Built on the open Model Context Protocol. Works with Claude, ChatGPT, Gemini, and any MCP client, across ROS 1 and ROS 2.

  • Plug and play

    Install in minutes with no changes to your robot code. Add a rosbridge node and connect. Real robots and simulators alike.

Running in five minutes

Open-source, Apache 2.0. No changes to your robot code. Just add a rosbridge node and connect.

Read the docsView on GitHub
  1. 01

    Install

    One command with pipx. Modules download on first run.

  2. 02

    Run

    Start the server. It opens a secure Cloudflare tunnel to your robot.

  3. 03

    Connect

    Point Claude, ChatGPT, or any MCP client at it and start commanding.

Proven in production

RobotMCP orchestrates multi-vendor warehouse robots today: a Contoro unloader and a Doosan palletizer, coordinated by an AI agent through one certified interface.

Talk to us about your fleet
300,000+
boxes unloaded in production
99.5%
AI success rate
0
safety incidents