MCP ServerOfficialHTTP

Hugging Face MCP Server

Hugging Face ships an official hosted MCP server at huggingface.co/mcp that gives any AI model search across the Hub — models, datasets, papers and Spaces — plus the ability to call selected Spaces as tools. Authenticate with a Hugging Face token and connect in seconds.

Hosted URL

https://huggingface.co/mcp

Suggested model

Claude Sonnet 4.5

Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.

Auth

Hugging Face access token from huggingface.co/settings/tokens, sent as an Authorization bearer header. A read token is enough for search.

What the Hugging Face MCP server does

How models use it and what it is built for.

The Hugging Face MCP server exposes the Hub as a set of tools an agent can reason over. A model can search for models matching a task and filter by library, licence or size, look up dataset cards, find papers, and inspect what a given repository contains. The more unusual capability is Spaces-as-tools: you choose Spaces from your MCP settings page and they become callable tools, so an agent can run an image generator or a transcription demo as a step in a larger task. Configure which tools and Spaces are exposed at huggingface.co/settings/mcp.

Tools the Hugging Face MCP server exposes

Typical tools an AI model can call. Exact names vary by version.

  • model_search — find models by task, library, licence or author
  • dataset_search — search datasets and read dataset cards
  • paper_search — look up papers indexed on the Hub
  • space_search — discover Spaces matching a description
  • model / dataset detail — inspect files, tags and metadata for a repo
  • user-selected Spaces — exposed as callable tools from your MCP settings

Example prompts to try

Copy any of these into MCP Agent Studio after connecting.

  • Find a small open-licence model for speech-to-text that runs on CPU.

  • What datasets exist for multilingual sentiment classification with more than 100k rows?

  • Compare the top three text embedding models by downloads and licence.

  • Search for recent papers on tool-use benchmarks for language models.

Models on MCP Playground

This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.

Default pick for Hugging Face

Claude Sonnet 4.5

Sonnet 4.5 handles multi-criteria search well and is good at narrowing a broad Hub query into a shortlist with reasons rather than dumping results.

Check an AI agent can actually use the Hugging Face MCP server

Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.

Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.

Run evals

Try the Hugging Face MCP server in your browser

Open MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.

Open Agent Studio

Hugging Face MCP server — FAQ

Common questions about connecting, scoping and using it safely.

What is the Hugging Face MCP server?

It is the official MCP server from Hugging Face, hosted at huggingface.co/mcp. It gives AI models search across the Hub — models, datasets, papers and Spaces — and lets you expose selected Spaces as callable tools for an agent to use mid-task.

Do I need a Hugging Face token?

Yes for the hosted endpoint. Create one at huggingface.co/settings/tokens and send it as an Authorization bearer header. A read-scoped token covers search; you only need write scope if you intend to push to repositories.

What does "Spaces as tools" mean?

You pick Spaces from huggingface.co/settings/mcp and they become tools your agent can call. That lets a model run a hosted demo — an image generator, a transcriber, a classifier — as one step inside a longer task, rather than you opening the Space by hand.

Can I run the Hugging Face MCP server locally?

Yes. The hf-mcp-server package can be run locally with npx or Docker if you would rather not use the hosted endpoint, which is useful when you want tighter control over which tools are exposed.

How do I test the Hugging Face MCP server in a browser?

Paste https://huggingface.co/mcp and your token into the MCP Server Tester to list the tools it exposes, or use MCP Agent Studio to give a real model a search task and watch each tool call and response — no local setup.

Other MCP servers

More on MCP Playground

Hugging Face MCP Server — AI Agent for Models and Datasets