Summary This project, lifemate-ai/embodied-claude, equips LLMs like Claude with physical embodiment—enabling perception (sight, hearing), voice, and mobility using affordable hardware. It transforms AI from passive text processors into active agents that interact with and remember the physical world, fostering novel human-AI interactions and advancing embodied AI research.

How It Works A modular system of "MCP Servers" provides distinct physical capabilities. Components like wifi-cam-mcp (vision/neck), tts-mcp (voice), and mobility-mcp (robot vacuum integration) leverage inexpensive hardware (e.g., ~$30 Wi-Fi cameras). This design prioritizes essential functions, allowing AI to "see for itself" and actively engage with its environment, rather than passively receiving data.

Quick Start & Requirements

• Primary Install: Clone the repo (git clone https://github.com/kmizu/embodied-claude.git), then set up individual MCP servers using uv sync within their directories. Configuration is managed via .mcp.json and .env files.
• Prerequisites:
  • Hardware: USB webcam (optional), Wi-Fi PTZ camera (e.g., TP-Link Tapo C210/C220), NVIDIA GPU (8GB+ VRAM for local Whisper), Tuya-compatible robot vacuum (optional).
  • Software: Python 3.10+, uv, ffmpeg 5+, OpenCV, Pillow, local Whisper, ElevenLabs API key (optional), VOICEVOX (optional), go2rtc, mpv/ffplay.
• Links: Repository: https://github.com/kmizu/embodied-claude.git

Highlighted Details

• Affordable Embodiment: Achieves AI embodiment using consumer-grade hardware, with core components costing approximately $30-$40.
• Modular MCP Architecture: Decouples physical capabilities (vision, hearing, voice, memory, mobility) into distinct micro-services.
• "Self-Seeing" AI: Enables LLMs to actively perceive their environment, moving beyond passive data consumption.
• Optional Mobility: Integrates with Tuya-enabled robot vacuums for physical navigation.
• Advanced Memory Systems: Includes modules for long-term, visual, episodic memory, and Theory of Mind (ToM).

Maintenance & Community Notable contributions are acknowledged from Rumia-Channel (ONVIF support) and fruitriin (interoception hook). claude-code-webui by sugyan is used for remote operation. No explicit community channels or detailed roadmap are provided.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive MIT license allows for commercial use and integration into closed-source projects.

Limitations & Caveats The system-temperature-mcp module is non-functional within WSL2 environments due to hardware access limitations. Autonomous action features require careful privacy considerations and manual cron job setup. Robot vacuum integration depends on Tuya device compatibility and 2.4GHz Wi-Fi.