This profile is part of our MCP server directory, within The MCP Guide. Profiles are checked against each vendor's own documentation. Need a server built? See MCP development.
The Hugging Face MCP server connects an AI app to the Hugging Face Hub. It can search models, datasets, Spaces and papers, search the docs, run jobs and call community AI apps built with Gradio. Hugging Face hosts it at https://huggingface.co/mcp and calls it the "official Hugging Face MCP Server."
Checked against: Hugging Face MCP Server docs on 28 October 2026. eSEOspace did not install or test this server.
Key Takeaways
- Hugging Face's official server is hosted at
https://huggingface.co/mcp, with open-source code you can also run locally. - The core built-in tool is
hf_fs, which navigates the Hub and runs semantic searches of docs and Spaces. - Extra tools you switch on in your MCP settings can create repos, run jobs and use sandboxes.
- You can add community Gradio Spaces as tools, or let the assistant discover them with "Dynamic Spaces."
- Sign in through your client with OAuth (
?login) or send a Hugging Face token as a bearer header.
At a glance
| Field | Detail |
|---|---|
| Publisher | Hugging Face |
| Official? | Yes |
| Hosting | Both: remote https://huggingface.co/mcp and local packages @llmindset/hf-mcp-server / Docker |
| Transport | Streamable HTTP (stateless JSON mode); stdio locally |
| Sign-in | OAuth or Hugging Face token |
| Read or write | Read by default; optional tools write (repos, jobs, sandboxes) |
| Price | No separate MCP price listed |
| Source code | github.com/huggingface/hf-mcp-server |
| Status | Not stated; docs say experimental tools are added from time to time |
What it can do
The docs list what the server lets an assistant do:
- Search and explore models, datasets, Spaces, papers and more.
- Search the Hugging Face documentation in plain language.
- Schedule and run Jobs and use sandboxes.
- Run community tools through MCP-compatible Gradio apps hosted on Spaces.
The docs say "Most Hub tasks can be efficiently completed with the built in hf_fs tool." Extra tools are switched on from your MCP settings page:
| Tool | What the docs say it does |
|---|---|
| Contribute Repos | Create and write files to those created repositories |
| Sandboxes | Create and use sandboxes, including file management |
| Run and Manage Jobs | Run, monitor and schedule jobs on Hugging Face infrastructure |
The docs name these extra tools by label, not by MCP tool name.
Spaces as tools. Gradio apps expose their functions as tools. You can add Spaces in your settings, or turn on Dynamic Spaces so the assistant finds and calls them at runtime. Remove Embedded Images strips images from Gradio results.
How to connect it
The docs recommend the settings page. Open it while logged in, pick your client, and copy the generated config. They say to use it "rather than writing config by hand."
The GitHub README gives this Claude Code command for OAuth sign-in:
claude mcp add hf-mcp-server -t http https://huggingface.co/mcp?loginOr with a token:
claude mcp add hf-mcp-server \
-t http https://huggingface.co/mcp \
-H "Authorization: Bearer <YOUR_HF_TOKEN>"For Claude Desktop or claude.ai, the README says to add "Hugging Face" from the connectors gallery. It also has steps for Gemini CLI, VS Code and Cursor.
To run locally, it gives npx @llmindset/hf-mcp-server for stdio and a Docker image. Add ?no_image_content=true to the URL to drop image blocks from Gradio servers.
Security notes
Write tools are opt-in. Repo creation, jobs and sandboxes are extra tools you choose in settings. The README says repository creation, Jobs, dynamic Space and sandbox tools "require Hugging Face authentication." Leave them off unless you need them.
Tokens. If you use a token rather than OAuth, give it the narrowest permissions Hugging Face allows and keep it out of shared config files. See MCP secret leakage.
Community Spaces. Gradio Spaces are built by community developers. A Space can return any text, and Dynamic Spaces lets the assistant pick ones you have not reviewed. Treat Space output as untrusted, as described in prompt injection through MCP tool results.
Jobs use compute. Jobs run on Hugging Face infrastructure under your account. Keep approvals on for them. The MCP security center has the full checklist.
Alternatives
- ElevenLabs MCP server: voice and audio agents rather than model discovery.
- Exa MCP server or Perplexity's MCP server: general web and paper search.
- Firecrawl MCP server: its research tools search papers from arXiv and PubMed.
More on our MCP server profiles hub.
Next step
You can also publish your own MCP server as a Gradio Space. If you want help building one for your product, our MCP development team can scope it.
Frequently asked questions
Is the Hugging Face MCP server free?
The docs do not list a separate MCP price. Jobs and sandboxes run on Hugging Face infrastructure under your account, so check your plan before using them.
What is hf_fs?
It is the main built-in tool. The docs say it lets the assistant navigate the Hub, including semantic searches of documents and Spaces.
How do I add a Gradio Space as a tool?
Find a Space with MCP support, then add it in your Hugging Face MCP settings. Restart or refresh your client so it picks up the new tools.
Can I run it myself?
Yes. The README documents npx @llmindset/hf-mcp-server for stdio, an HTTP mode and a Docker image. The hosted server is the simpler option for most people.
