Submit Log in

Hugging FaceMCP Server

by Hugging Face

Released Oct 22, 2025

6 views

v0.4.10

MIT

Repository Docs npm

Featured Developer Tools

Summary

Hugging Face's official remote MCP server — search and inspect models, datasets and Spaces, browse the Hub filesystem over hf:// URIs, and turn Gradio Spaces into callable tools.

Features

hub_repo_search: one query interface across models, datasets and Spaces
hub_repo_details: metadata for one or more repos with type auto-detection
hf_fs: ls/cat/find/stat/search over hf:// URIs across the whole Hub
hf_whoami: inspect the active account, orgs and credential scope
Expose chosen Gradio Spaces as generated, callable tools
OAuth (?login) or HF token auth; anonymous access is rate-limited
Self-hostable over stdio or stateless streamable HTTP via npx or Docker

Installation

Set up this MCP server in your favorite AI agent — copy a ready-made configuration below.

Any MCP-compatible agent

Most agents (Claude, Cursor, Windsurf, VS Code, and more) read a standard mcpServers configuration.

Open your agent's MCP configuration file.
Merge the snippet below into it.
Restart the agent — the "Hugging Face" tools will be available.

{
  "mcpServers": {
    "huggingface": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://huggingface.co/mcp?login"
      ]
    }
  }
}

Claude Code

Add the server with a single CLI command.

claude mcp add huggingface -- npx -y mcp-remote https://huggingface.co/mcp?login

Claude Desktop

Add the server to claude_desktop_config.json (Settings → Developer → Edit Config).

{
  "mcpServers": {
    "huggingface": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://huggingface.co/mcp?login"
      ]
    }
  }
}

Cursor

Or configure the HTTP URL directly: {"huggingface": {"url": "https://huggingface.co/mcp", "headers": {"Authorization": "Bearer <YOUR_HF_TOKEN>"}}}

{
  "command": "npx",
  "args": [
    "-y",
    "mcp-remote",
    "https://huggingface.co/mcp"
  ]
}

VS Code

{
  "command": "npx",
  "args": [
    "-y",
    "mcp-remote",
    "https://huggingface.co/mcp?login"
  ]
}

Self-host over stdio

{
  "command": "npx",
  "args": [
    "@llmindset/hf-mcp-server"
  ],
  "env": {
    "HF_TOKEN": "hf_xxxxxxxx"
  }
}

Self-host in stateless streamable HTTP mode; management dashboard at /metrics

{
  "command": "docker",
  "args": [
    "run",
    "--rm",
    "-p",
    "3000:3000",
    "ghcr.io/evalstate/hf-mcp-server:latest"
  ]
}

Description

The Hugging Face MCP server puts the Hub itself inside an agent. Point a client at https://huggingface.co/mcp and the model stops guessing which checkpoint exists: it can search models, datasets and Spaces through one query interface, pull details for specific repos, and read files straight out of a repo without you downloading anything first.

The built-in tools (what an anonymous connection sees today, at server version 0.4.10):

hub_repo_search — one search surface across models, datasets and Spaces, or aggregated across all three.
hub_repo_details — full metadata for one or more repos, with the type auto-detected.
hf_fs — a filesystem over the Hub: ls, cat, find, stat, attach and search against hf:// URIs, rooted at models, datasets, Spaces and buckets.
hf_whoami — which account and organisations the current credential can see, useful for debugging a permissions problem before blaming the tool.

Gradio Spaces as tools. The interesting part is what you add to that. At huggingface.co/settings/mcp you choose which Spaces to expose, and the server generates tool aliases for them — so an image model, a transcription demo or any other Gradio app running on the Hub becomes something the agent can call directly. That is how one endpoint turns into thousands of task-specific tools without any of them being packaged or shipped.

Connecting. It is a remote streamable-HTTP server. Use https://huggingface.co/mcp?login for a browser OAuth flow, or https://huggingface.co/mcp with an Authorization: Bearer hf_... header for a token. Anonymous use works but is rate-limited, and the server says so in its own instructions. Claude Code, Claude Desktop and claude.ai (via the connector gallery), Cursor, VS Code, Gemini CLI and any other MCP client with HTTP transport are supported. You can also self-host it — npx @llmindset/hf-mcp-server for stdio, or the ghcr.io/evalstate/hf-mcp-server image for stateless streamable HTTP — which is the route to take if you need DISABLE_TOOLS, host allowlists or your own metrics dashboard.

MIT licensed, and the reference implementation is developed in the open.

Related MCP Servers

MCP: Engram

by Gentleman Programming

New

Persistent memory for coding agents as a single Go binary — SQLite and FTS5 behind MCP, an HTTP API, a CLI and a TUI, with no Node, Python or Docker required.

Developer Tools

MCP: agent-device

by Callstack

New

Give a coding agent a live feedback loop on real mobile, TV and desktop apps — inspect, tap, type and capture evidence on iOS, Android and HarmonyOS over MCP or a CLI.

Developer Tools

MCP: Weights & Biases MCP Server

by Weights & Biases

New

The official W&B MCP server: query experiment runs, Weave LLM traces, artifacts and registries in natural language, and write findings back as a W&B report.

Developer Tools

MCP: Replicate MCP Server

by Replicate

Featured

Replicate's official MCP server: search thousands of hosted models, read their schemas, and run predictions on image, video, audio and language models from inside an agent.

Developer Tools

Browse all MCP servers →