Hugging Face

This MCP server provides comprehensive access to Hugging Face's ecosystem through multiple transport protocols (STDIO, SSE, HTTP) with both built-in tools for searching spaces, models, datasets, and papers, plus dynamic integration with Gradio-based MCP servers hosted on Hugging Face Spaces. Built by Hugging Face using TypeScript with a React-based web interface for tool management, it offers semantic search across the Hub's content, detailed information retrieval for specific resources, space duplication capabilities, and a proxy architecture that automatically discovers and registers remote tools from configured Gradio endpoints. The implementation includes intelligent tool selection strategies with bouquet-based filtering, transport metrics tracking, and support for both stateful session-based and stateless JSON modes, making it valuable for AI assistants that need to discover ML resources, manage Hugging Face workflows, or dynamically extend capabilities by connecting to community-built tools hosted as Spaces.

Connection

• streamable_http · auth: api_key · free_tier — https://huggingface.co/mcp

Source

Repository: https://github.com/huggingface/hf-mcp-server