A set of zero-dependency MCP servers to coordinate shared hardware usage across multiple agent instances (Gemini CLI, Claude Code, etc.) on a single host.

• GPU Coordination: Manages atomic locks for NVIDIA GPUs.
• Device Coordination: Manages atomic locks for Flutter mobile/web devices.

Install directly from GitHub as a native extension:

gemini extensions install https://github.com/tianhaoz95/mcp

This handles cloning and tool registration automatically.

1. Clone the repository:

   git clone https://github.com/tianhaoz95/mcp
   cd mcp

2. Register the tools:

   For Claude Code (Global):

   # GPU Coordination
   claude mcp add gpu-coordination python3 $(pwd)/tools/gpu-coordination/gpu_mcp_server.py --env GPU_COUNT=8 --scope user
   
   # Device Coordination
   claude mcp add device-coordination python3 $(pwd)/tools/device-coordination/device_mcp_server.py --scope user

   Note: To limit these tools to a specific project, omit the --scope user flag.

   For Gemini CLI (Manual):

   gemini mcp add gpu-coordination python3 $(pwd)/tools/gpu-coordination/gpu_mcp_server.py --env GPU_COUNT=8
   gemini mcp add device-coordination python3 $(pwd)/tools/device-coordination/device_mcp_server.py

• Zero Dependency: Only requires Python 3 (standard library).
• Atomic Locking: Prevents resource contention between agents.
• Shared State: All agents on the same host share the same inventory.
• Auto-Release: Prevents deadlocks if an agent crashes.