Hugging Face MCP server
Hugging Face's official MCP server: search and explore models, datasets, Spaces, papers, and docs from your AI assistant.
The Hugging Face MCP server is Hugging Face's official integration that connects an MCP-compatible assistant directly to the Hugging Face Hub. Instead of switching to the website, the agent can semantically search models, datasets, Spaces, and research papers, pull detailed metadata for any repository, search the Hugging Face documentation in natural language, and run or manage Jobs on Hugging Face infrastructure. Each built-in tool can be toggled on or off from your MCP settings, and the server also dynamically integrates community Gradio Spaces, so any MCP-compatible Space you select becomes an additional callable tool, including image generation, transcription, and other on-demand AI apps.
The canonical deployment is the hosted remote endpoint at https://huggingface.co/mcp, which speaks both Streamable HTTP and stdio and authenticates with your Hugging Face access token sent as a bearer Authorization header. For local or self-hosted use, the open-source server is published on npm (@llmindset/hf-mcp-server and its HTTP variant) and as a prebuilt Docker image, with the same token-based auth. The configuration page at huggingface.co/settings/mcp generates the exact client snippet and lets you choose which tools and Spaces are exposed.
Quick install
Copy-paste configs are provided for all 8 supported clients. Pick your client below.
Add to ~/.claude.json
{
"mcpServers": {
"huggingface": {
"type": "http",
"url": "https://huggingface.co/mcp",
"headers": {
"Authorization": "Bearer <HF_TOKEN>"
}
}
}
}claude mcp add --transport http huggingface https://huggingface.co/mcpAvailable tools
| Tool | Description |
|---|---|
| space_search | Spaces semantic search: find the best AI apps via natural-language queries (e.g. TTS, ASR, OCR). |
| paper_search | Papers semantic search: find ML research papers via natural-language queries. |
| model_search | Search models with filters for task, library, and other metadata. |
| dataset_search | Search datasets with filters for author, tags, and other metadata. |
| hf_doc_search | Documentation semantic search across Hugging Face libraries for guides, API references, and tutorials. |
| hf_jobs | Run, monitor, and schedule jobs on Hugging Face infrastructure. |
| hub_repo_details | Get detailed information about models, datasets, and Spaces, optionally including README content. |
Required configuration
- HF_TOKENRequired
Hugging Face access token used for the bearer Authorization header on the remote endpoint or the stdio server. Required.
What you can do with it
Find the right model or dataset without leaving the agent
Use model_search and dataset_search with task and metadata filters, then hub_repo_details to inspect a candidate repository's card and README before wiring it into your code.
Discover and run community AI apps
Search Spaces with space_search to find an app that does what you need (transcription, image generation, OCR), then call MCP-compatible Gradio Spaces directly as tools to get results back inline.
FAQ
- Is it free?
- Yes. The MCP server is open source and free, and you only need a free Hugging Face account and access token. Underlying paid features (such as Jobs compute or gated models) are governed by your own Hugging Face plan.
- Does it support remote/OAuth?
- Yes. The canonical deployment is the hosted remote endpoint at https://huggingface.co/mcp over Streamable HTTP, authenticated with your Hugging Face access token as a bearer header. A self-hostable stdio/HTTP server and Docker image are also published.