DEV Community

curatedmcp for CuratedMCP

Posted on • Originally published at curatedmcp.com

Hugging Face MCP: Search 500k+ Models and Run Inference Directly From Claude

Install guide and config at curatedmcp.com

Hugging Face MCP: Search 500k+ Models and Run Inference Directly From Claude

The Hugging Face MCP server connects your AI agent directly to the world's largest open-source model repository. Instead of leaving your Claude or Cursor session to browse Hugging Face, search models, run inference, explore datasets, and deploy Spaces without context-switching.

What It Does

This server unlocks five core capabilities:

Model discovery across 500,000+ models filterable by task, framework, and performance metrics. Find the right model for your use case without manual browsing.

Inference execution lets you run predictions through any model via the Hugging Face Inference API—pass text, images, or audio and get results back in your agent's context.

Dataset access to 150,000+ community datasets with schema inspection and preview data, useful for training, fine-tuning, or research workflows.

Space management to deploy and monitor Gradio/Streamlit applications directly from your agent.

Leaderboard comparisons so you can check Open LLM Leaderboard rankings and model benchmarks programmatically.

You get model cards, tokenizer configs, community discussions, and weight downloads all accessible without switching tabs.

How to Install

Install via npm:

npx -y @huggingface/mcp
Enter fullscreen mode Exit fullscreen mode

Add to your Claude Desktop config:

{
  "mcpServers": {
    "hugging-face-mcp": {
      "command": "npx -y @huggingface/mcp"
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

After installation, restart Claude Desktop and the server will appear in your MCP options.

Real-World Use Cases

  • Model selection: Ask Claude, "Find me the best open-source sentiment analysis model under 1B parameters with the highest F1 score on the leaderboard." Get ranked results with links and stats instantly.

  • Inference pipelines: "Run this customer review through the Mistral-7B model and give me the sentiment score." Execute inference without writing API calls.

  • Data discovery: "What datasets are available for training a code completion model? Show me their sizes and schemas." Explore training data options while brainstorming architecture.

Full install guides for Claude Desktop, Cursor, Windsurf, and more at CuratedMCP.

Top comments (0)