Hugging Face Hub MCP Server

A Model Context Protocol (MCP) server that provides access to the Hugging Face Hub API, allowing you to search and retrieve information about models, datasets, and their metadata.

Features

• Models API: Search models, get detailed model info, and retrieve model tags
• Datasets API: Search datasets, get dataset info, access parquet files, and Croissant metadata
• Full Hub Integration: Access all public repositories and metadata from Hugging Face Hub

Tools

Models

• hf_list_models - List and search models with filtering options
• hf_get_model_info - Get detailed information about a specific model
• hf_get_model_tags - Get all available model tags

Datasets

• hf_list_datasets - List and search datasets with filtering options
• hf_get_dataset_info - Get detailed information about a specific dataset
• hf_get_dataset_parquet - Get parquet file information for datasets
• hf_get_croissant - Get Croissant metadata for datasets
• hf_get_dataset_tags - Get all available dataset tags

Installation

npm install
npm run build

Usage

STDIO (for Claude Desktop)

npm run dev:stdio

HTTP Server

npm run dev:shttp
# or specify port
npm start --port 3003

Configuration

The server uses the Hugging Face Hub public API endpoints. No authentication is required for public repositories.

Environment variables:

• HF_BASE_URL: Base URL for Hugging Face API (default: https://huggingface.co)
• PORT: HTTP server port (default: 3003)

Examples

Search for text classification models:

{
  "name": "hf_list_models", 
  "arguments": {
    "filter": "text-classification",
    "limit": 10
  }
}

Get information about a specific model:

{
  "name": "hf_get_model_info",
  "arguments": {
    "repo_id": "microsoft/DialoGPT-medium"
  }
}

Search for datasets by author:

{
  "name": "hf_list_datasets",
  "arguments": {
    "author": "huggingface",
    "limit": 5
  }
}