M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
H

Hf MCP Server

Official
Connect your LLM to the Hugging Face Hub and thousands of Gradio AI Applications.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #158 of 438
Stars
★ 303 Very popular
Transport
stdio (local process) · Streamable HTTP
Runtime
Node.js · Docker · Hosted (no local runtime)
Credentials
Optional API key
License
MIT
Last commit
57FMRS · C

Official Hugging Face MCP server, feature-rich, supports multiple clients and transports, suitable for integrating Hugging Face ecosystem.

Strongest · Setup experience 14/20 Weakest · Reliability 10/20

Reliability
10/20
Security and permissions
11/20
Maintenance
10/20
Documentation
12/20
Setup experience
14/20
Why each score
Reliability 10/20
Evidence: the README documents STDIO and StreamableHTTP/JSON transports, local startup via npx/docker, and several environment variables (timeout, host allowlist, tool disabling), making a happy path plausible. Deduction: no actual tool list is provided, and there is no verifiable CI workflow or committed test suite, so the init/tools-list handshake and declared tool behavior cannot be confirmed from the source material.
Security and permissions 11/20
Evidence: the README requires an Authorization Bearer token for HTTP transports, supports DEFAULT_HF_TOKEN/HF_TOKEN for local STDIO, and mentions outbound checks, ALLOW_INTERNAL_ADDRESS_HOSTS, DISABLE_TOOLS, and proxy-tool source constraints, showing security awareness. Deduction: there is no documentation of least-privilege/read-only token usage, confirmation for side effects from Gradio calls, integrity verification of proxy-tool sources, or a security audit/disclosure channel, so the score is limited.
Maintenance 10/20
Evidence: the repository is under the Hugging Face organization, has an MIT license, is not archived, and the README shows ongoing build/Docker development. Deduction: commit frequency, release cadence, dependency update history, and issue response behavior are not visible from the README; the 13 open issues are not characterized, so this is scored as active but with governance/versioning gaps.
Documentation 12/20
Evidence: the README covers installation for Claude, Gemini, VSCode, and Cursor, plus local running, Docker, environment variables, proxy tools, and Skills, with generally clear examples. Deduction: it does not list the actual MCP tools or their parameters, and it lacks limits, cost, troubleshooting, and known-issue sections, forcing users to infer capabilities from external sources.
Setup experience 14/20
Evidence: the README provides concrete configuration snippets for mainstream clients, npx/docker startup commands, and clear prerequisites, making the setup path look straightforward. Deduction: there is no CI or test evidence proving these steps work out of the box; setup also depends on an HF account/token and, for the default remote URL, extra authentication and network configuration.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Developers wanting to integrate Hugging Face ecosystem into LLM tools
  • AI application builders needing access to models, datasets, and Gradio apps

Not for

  • Users not using Hugging Face services
  • Scenarios requiring long-lived connections or stateful MCP servers (STDIO does not expose skills)

Required permissions

  • Requires a Hugging Face token (HF_TOKEN) for authentication
  • HTTP transport requires Authorization: Bearer header
  • May access Hugging Face API and Gradio apps

Risks and side effects

  • User-supplied tokens may be misused
  • Proxy tools or external URLs may introduce security risks
  • Local loopback addresses are allowed by default; external access requires allowlist

Setup

Before you start

Runtime：Node.js · Docker · Hosted (no local runtime)

DEFAULT_HF_TOKEN optionalsecret Default token for local STDIO deployments; get one at https://huggingface.co/settings/tokens.
HF_TOKEN optionalsecret Fallback Hugging Face token used with stdio transport when DEFAULT_HF_TOKEN is not set.
PROXY_TOKEN optionalsecret Optional token used only for startup authentication while discovering PROXY_TOOLS_CSV schemas.
Other optional settings (10)
TRANSPORT optional Transport type: stdio or streamableHttpJson (Docker defaults to the latter).
MCP_ALLOWED_HOSTS optional Extra comma-separated Host allowlist for MCP/API routes; supports exact hosts and *.example.com wildcards.
HF_API_TIMEOUT optional Timeout in milliseconds for Hugging Face API requests, default 12500.
USER_CONFIG_API optional Optional URL for per-user MCP settings; built-in immutable defaults are used when unset.
ALLOW_INTERNAL_ADDRESS_HOSTS optional Comma-separated allowlist of trusted domains permitted to resolve internal/reserved DNS (exact hosts or * wildcards).
MCP_STRICT_COMPLIANCE optional Set to True to reject GET requests with 405 in JSON mode (default serves a welcome page).
DISABLE_TOOLS optional Comma-separated tool names to hide from tools/list and reject when called.
PROXY_TOOLS_CSV optional HTTPS URL or local file path defining Streamable HTTP proxy tool sources loaded at startup.
GRADIO_SKIP_INITIALIZE optional When true, Gradio MCP calls skip the initialize handshake and issue tools/call directly.
HF_SKILLS_DIR optional Local directory containing a prebuilt skills snapshot; defaults to /mnt/hf-skills/distribution/latest.
  1. For Claude Desktop or claude.ai, click the link in the README or add 'Hugging Face' from the connectors gallery.
  2. For Claude Code, run: claude mcp add hf-mcp-server -t http https://huggingface.co/mcp?login (or use a token).
  3. For Gemini CLI, run: gemini mcp add -t http huggingface https://huggingface.co/mcp?login.
  4. For VSCode, click the link in README or install from the MCP gallery; alternatively configure manually in mcp.json.
  5. For Cursor, click the install link in README or configure manually.
  6. You can also run locally via npx or Docker: npx @llmindset/hf-mcp-server or docker run --rm -p 3000:3000 ghcr.io/evalstate/hf-mcp-server:latest.

Check that it works

The client's tool list should show Hugging Face tools (such as hub_repo_search or hf_fs, unless hidden via DISABLE_TOOLS). Alternatively, run npx @llmindset/hf-mcp-server and check server status and MCP metrics at http://localhost:3000/.

Troubleshooting

  1. Check environment variables such as TRANSPORT and DEFAULT_HF_TOKEN
  2. Ensure network connectivity to Hugging Face services
  3. Inspect the management dashboard on port 3000 for status and metrics
  4. For HTTP requests, ensure the Authorization header contains a valid token
  5. If tool list is empty, check DISABLE_TOOLS or user configuration

Things to try

Once connected, you can ask your AI assistant things like:

  • Search the Hugging Face Hub for text-to-speech model repositories
  • Use hf_fs to list the files in a Hugging Face dataset repository
  • Find Gradio AI apps that can generate images
  • Get details about a Hugging Face dataset

Use cases

Interact with the Hugging Face Hub, e.g., search models, datasets, Spaces
Call AI functionalities through Gradio apps
Build AI workflows in MCP-enabled clients

Supported clients

Claude Desktop
Claude Code
Gemini CLI
VS Code
Cursor

Listed from the project's documentation, not tested by this site.

Overview

The official Hugging Face MCP Server provides MCP implementations of Hub API and Search endpoints, along with an MCP server and web application for deploying endpoints. It supports STDIO and stateless Streamable HTTP JSON transports, with tool selection configurable per request via user configuration API and query parameters.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 62.9k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision dce2d5f0f163 Data synced 2026-10-12 Read the FMRS scoring method