bui21x/huggingface-mcp-server

HuggingFace MCP Server

A Model Context Protocol (MCP) server that provides HuggingFace model integration for AI agents.

Features

• Model inference with GPU support
• Model caching for improved performance
• Multiple task support
• Custom parameter configuration
• Health monitoring with GPU status

Setup

1. Install dependencies:

pip install -r requirements.txt

2. Set environment variables:

HUGGINGFACE_TOKEN=your_token_here  # Optional

3. Run server:

uvicorn src.mcp_server:app --reload

API Endpoints

• POST /inference - Run model inference
• GET /models - List loaded models
• GET /health - Check server health

Example Usage

# Run inference
POST /inference
{
    "model_id": "gpt2",
    "task": "text-generation",
    "inputs": "Once upon a time",
    "parameters": {
        "max_length": 50,
        "temperature": 0.7
    }
}

MCP Integration

This server follows the MCP specification for tool integration with AI agents.