Hugging Face MCP Server: Search the Hub from Claude
The official Hugging Face MCP server lets Claude search models, datasets, Spaces and papers, and call Gradio apps as tools. Claude Code setup and use cases.
By Claudekit Team Published 2 min read
- Type
- MCP Server
- Pricing
- Freemium
- Repository
- github.com
- GitHub stars
- 308
- Website
- huggingface.co
The Hugging Face MCP Server is the official Model Context Protocol server for the Hugging Face Hub. It lets Claude search and explore models, datasets, Spaces and papers, and call community Gradio apps hosted on Spaces as tools. Results come back with metadata and links you can open on the Hub.
Who it is for
ML engineers and developers who want Claude to find the right model or dataset for a task, compare options, or run a Space, such as transcription or image generation, without leaving the chat.
Install
The README documents the Claude Code command. It uses the hosted endpoint and a browser login:
claude mcp add hf-mcp-server -t http https://huggingface.co/mcp?loginStart claude and follow the prompts to finish authentication. To use a token instead:
claude mcp add hf-mcp-server \
-t http https://huggingface.co/mcp \
-H "Authorization: Bearer <YOUR_HF_TOKEN>"Use a token with the narrowest scope that works, and never commit it. After installing, visit huggingface.co/settings/mcp to pick your tools and Spaces.
Use cases
- Model search: “Find quantized Qwen 3 models and compare their downloads and licenses.”
- Datasets: “Show datasets about weather time series.”
- Run a Space: find a Space that transcribes audio and call it as a tool.
- Papers: look up recent papers on a technique and link to them.
Pros
- Official, MIT licensed and hosted by Hugging Face.
- Documented Claude Code command with login or token.
- Extensible: add any MCP-compatible Gradio Space as a tool.
- Tool selection is managed in your Hub settings, not in local config.
Cons and caveats
- Community Spaces are third-party code. Check what a Space does before adding it.
- Some Spaces need GPU time or have usage limits on your account.
- Image results from Gradio apps can be large. Use
no_image_content=trueif you only need text.
Links
- Repo: huggingface/hf-mcp-server
- Docs: Hugging Face MCP Server
- Related: Hugging Face skills, Context7, what is MCP
Checked 24 Oct 2026 against the vendor repo, the Hugging Face docs and the MCP registry.
Frequently asked questions
Where do I choose which tools are enabled?
In your Hugging Face account at huggingface.co/settings/mcp. That page also lets you add MCP-compatible Gradio Spaces as extra tools.
Can I use a token instead of browser login?
Yes. The README shows adding the server with an Authorization: Bearer header carrying your Hugging Face token.
How do I stop Gradio apps from returning images?
Add ?no_image_content=true to the server URL to remove ImageContent blocks from Gradio servers.
What license is it under?
The repository is MIT licensed.