Skip to content

MCP setupData and AI

Hugging Face MCP Server

Search Hugging Face models, datasets, Spaces and papers from your agent through the official hosted MCP server and a read token.

npx workspacetemplates add huggingface-mcp --tool claude-code

Writes 1 file to your project root

.mcp.jsonmerge into your file11 lines · 191 BFile
{
  "mcpServers": {
    "huggingface": {
      "type": "http",
      "url": "https://huggingface.co/mcp",
      "headers": {
        "Authorization": "Bearer ${HF_TOKEN}"
      }
    }
  }
}

What it does

Hugging Face runs a hosted MCP server for the Hub. Once connected, the agent can look up models and datasets, find Spaces that do a given job, search papers, and query the Hub documentation in plain language.

No local package is involved: point the client at the server URL and send a Hugging Face access token in an Authorization header. Extra tools such as repository creation, jobs and sandboxes can be switched on or off in your account's MCP settings, so you decide how much the agent can do.

Example

you › Find three text-to-speech models under 1B parameters with a permissive license.

agent › Three candidates with download counts and license: model A (Apache-2.0), model B (MIT), model C (Apache-2.0), each with its Hub link.

Questions

Which token should I use?
A fine-grained or read token is enough for search. Avoid a write token unless you want repository tools.
Where do I choose the tools?
In the MCP page of your Hugging Face account settings, where you can also add community Spaces as tools.
Is the setup the same for every client?
The settings page generates the snippet for your client. This entry uses the plain URL with a bearer header.
  • Apify MCP Server

    MCP setupData and AI

    Free
  • Chroma MCP Server

    MCP setupData and AI

    Free
  • ClickHouse MCP Server

    MCP setupData and AI

    Free
  • Firecrawl MCP Server

    MCP setupData and AI

    Free