xlerobot-mcp

# XLeRobot Servo MCP Server An MCP server that exposes the XLeRobot servo and wheel controls from RoboCrew's `servo_controls.py` to any MCP-compatible AI client. This server stays close to the upstream `ServoControler` API, but adds: - a standard MCP tool surface - a `mock` backend for development without hardware - camera observation through an MCP image tool - a `get_robot_state` tool and `robot://state` resource for structured state - both `stdio` and `streamable-http` transports ## TL;DR Install `uv`, then start the mock server instantly without cloning the repo: ```bash uvx xlerobot-mcp --backend mock ``` Run hardware mode the same way (CPU-only torch wheels): ```bash uvx --from "xlerobot-mcp[hardware]" --torch-backend cpu \ xlerobot-mcp \ --backend hardware \ --right-arm-wheel-usb /dev/ttyUSB0 \ --camera-index-or-path 0 ``` For Codex: ```bash codex mcp add xlerobot -- uvx xlerobot-mcp --backend mock ``` ## What It Exposes The server wraps the same control areas as the upstream RoboCrew module: - visual observation: `get_camera_image` - wheel motion: `move_forward`, `move_backward`, `turn_left`, `turn_right`, `strafe_left`, `strafe_right`, `stop_wheels` - head motion: `turn_head_yaw`, `turn_head_pitch`, `turn_head_to_vla_position`, `reset_head_position` - arm pose control: `set_arm_position`, `read_arm_present_position`, `save_arm_position`, `set_saved_position` - torque management: `enable_torque`, `disable_torque` - state inspection: `get_robot_state` ## Quick Start Use Python 3.11+. ```bash uv sync uv run xlerobot-mcp --backend mock ``` That starts a `stdio` MCP server in mock mode, which is the easiest way to connect from Codex, Claude Desktop, Cursor, Goose, or any other client that launches MCP servers as subprocesses. Once the package is published to PyPI, users can skip cloning the repo and run it directly with `uvx`: ```bash uvx xlerobot-mcp --backend mock ``` For real servo control from PyPI, include the hardware extra and…