The official Hugging Face MCP Server connects your LLM to the Hugging Face Hub and thousands of Gradio AI applications. It is available as a hosted endpoint at https://huggingface.co/mcp and can also be run locally over STDIO or stateless Streamable HTTP via npx or Docker. Tools and Spaces are configured at huggingface.co/settings/mcp.
What it does
Where Hugging Face MCP Server runs and what it adds, from its documentation
Exposes Hugging Face Hub API and search endpoints as MCP tools
Connects to Gradio Spaces as tools
Offers a bouquet=openai option for Hub filesystem, repo search/details/creation, dynamic Space, Jobs and sandbox tools
Runs over STDIO or Streamable HTTP (JSON mode) at /mcp
Provides a metrics web interface at /metrics when self-hosted
When to use it
Good fit
Searching Hub models, datasets and Spaces from an agent
Letting an agent call Gradio-based AI apps as tools
Self-hosting an MCP endpoint for the Hub in your own infrastructure
Not for
Setups where sending a Hugging Face token to the hosted endpoint is unacceptable (run locally instead)
How to install Hugging Face MCP Server
You need: Hugging Face account or token for authenticated tools; Node.js (npx) or Docker for local runs
Claude Code: claude mcp add hf-mcp-server -t http https://huggingface.co/mcp?login
Start claude and follow the instructions to authenticate
https://huggingface.co/mcp; add ?login to use the login flow. source
How do I add the Hugging Face MCP Server to Claude Code?
Run claude mcp add hf-mcp-server -t http https://huggingface.co/mcp?login, then start claude and complete authentication. source
Can I run the Hugging Face MCP Server locally?
Yes: npx @llmindset/hf-mcp-server for STDIO, npx @llmindset/hf-mcp-server-http for Streamable HTTP, or the ghcr.io/evalstate/hf-mcp-server Docker image. source
How do I limit which tools the Hugging Face MCP Server exposes?
Self-hosted, set DISABLE_TOOLS to a comma-separated list of tool names; hosted, configure tools at huggingface.co/settings/mcp. source