Repository files navigation

TheROS MCP Client is a reference implementation of a Model Context Protocol (MCP) client, designed to connect directly withros-mcp-server.

Instead of using a Desktop LLM client, it acts as a bridge that integrates an LLM, enabling natural-language interaction with any ROS or ROS2 robot.

robot-mcp-client implements the LLM-side of the MCP protocol.

It can:

• Connect to aros-mcp-server over MCP (stdio or HTTP).
• Send natural language queries or structured requests to the robot without the need to integrate it with a Desktop LLM client
• Stream back feedback, sensor data, or responses from the server.
• Integrate with multiple LLM providers including OpenAI, Anthropic, Gemini, and Groq (Llama, GPT-OSS).

In short, it lets you run an MCP-compatible client that speaks to robots via the MCP interface — useful for testing, local reasoning, or autonomous AI controllers.

• Implements MCP client specification — plug-and-play with the ROS MCP server.

• ROS-aware LLM interface — specialized prompts and handlers for robotics tasks.

• Supports bidirectional streaming — send commands, receive real-time topic feedback.

• LLM integration ready — use Gemini, Anthropic, or Ollama APIs as reasoning engines.

• Offline-capable — works entirely within local or LAN environments.

The MCP client is version-agnostic (ROS1 or ROS2).

• ROS or ROS2 running withrosbridge
• Activeros-mcp-server instance

1. Clone the repository

git clone https://github.com/robotmcp/robot-mcp-client.gitcd robot-mcp-client

2. Install dependencies

uv sync# or pip install -e .

3. Minimal setup for custom MCP client

./setup.sh# Then edit .env and set:# - ROS_MCP_SERVER_PATH=/absolute/path/to/ros-mcp-server# - LLM_PROVIDER=gpt-oss|openai|anthropic|ollama

4. Run the base client

uv run clients/baseclient.py

5. Startrosbridge on the target robot

ros2 launch rosbridge_server rosbridge_websocket_launch.xml

robot-mcp-client/├── .github/│   └── workflows/│       └── test-setup.yml    # CI for cross-platform setup├── clients/│   ├── baseclient.py         # Multi-LLM MCP client (LangGraph)│   ├── llm_store.py          # LLM provider configuration│   └── gemini_live/          # Gemini Live client│       ├── gemini_client.py  # Main client script│       ├── mcp.json          # MCP server configuration│       ├── setup_gemini_client.sh  # Automated setup│       └── README.md         # Detailed setup guide├── .env                      # Environment config (not tracked)├── setup.sh                  # Cross-platform setup script├── pyproject.toml            # Python dependencies└── README.md                 # This file

The project includes multiple LLM client implementations:

Multi-provider LLM client with support for:

• OpenAI: GPT-4.1
• Anthropic: Claude Sonnet 4.5
• Google Gemini: Gemini 2.5 Flash Lite
• Groq (open-source models):
  • Llama 4 Scout 17B
  • Llama 3.1 8B Instant
  • Llama 3.3 70B Versatile
  • OpenAI GPT-OSS 120B

Configuration: SetLLM_PROVIDER in.env (see setup.sh)

Usage: SetLLM_PROVIDER=<keyword> in your.env file (e.g.,LLM_PROVIDER=llama-scout)

• Full-featured Google Gemini integration
• Text-only mode optimized for WSL
• Real-time interaction with ROS robots
• Automated setup withsetup_gemini_client.sh

# Multi-provider base client./setup.sh# Edit .env: set ROS_MCP_SERVER_PATH and LLM_PROVIDERuv run clients/baseclient.py# Gemini Live clientcd clients/gemini_live./setup_gemini_client.shuv run gemini_client.py

We welcome community PRs with new client implementations and integrations!

We love contributions of all kinds:

• Bug fixes and documentation updates
• New features (e.g., Action support, permissions)
• Additional examples and tutorials

Check out thecontributing guidelines and see issues taggedgood first issue to get started.

This project is licensed under theApache License 2.0.

aggedgood first issue to get started.