Hugging Face MCP Server — Hosted for Any AI Agent

mcp-huggingface — Hugging Face MCP Server

Search models, datasets, and Spaces on Hugging Face — browse trending AI models, check downloads, and run inference from any agent.

Give your AI agents access to the largest open-source AI model hub. Search 900K+ models and 200K+ datasets, inspect model cards and configs, browse Spaces, list repository files, and run serverless inference — all through natural language.

Live endpoint: https://mcp.aerostack.dev/s/aerostack/mcp-huggingface

---

What You Can Do

• Search models by keyword, task, library, or author
• Get model details with tags, downloads, safetensors info, and model card
• Search datasets by keyword or author with download stats
• Browse Spaces (apps/demos) with SDK info and status
• Run serverless inference on any supported model
• List files in model repositories with LFS details

Available Tools

Tool	Description
_ping	Verify HF connectivity and show authenticated user
search_models	Search models by keyword, task, library, author
get_model	Get model details — tags, config, downloads, model card
search_datasets	Search datasets by keyword or author
get_dataset	Get dataset details — card, splits, features, downloads
search_spaces	Search Spaces (apps) by keyword
run_inference	Run inference on a model via serverless Inference API
list_model_files	List files in a model repo with sizes and LFS info

Configuration

Variable	Required	Description	How to Get
HUGGINGFACE_TOKEN	Yes	Hugging Face User Access Token with read permissions	huggingface.co → Settings → Access Tokens → New token → select "Read" or "Fine-grained"

For inference: Your token needs Inference permission. Some models (gated) require accepting terms on the model page first.

Quick Start

Add to Aerostack Workspace

1. Go to aerostack.dev → Your Project → MCPs
2. Search for "Hugging Face" and click Add to Workspace
3. Add HUGGINGFACE_TOKEN under Project → Secrets

Example Prompts

"Search for the most popular text-generation models"
"Show me details about meta-llama/Llama-3-8B-Instruct"
"Find datasets related to code generation"
"What are the trending Spaces right now?"
"Run inference on meta-llama/Llama-3-8B-Instruct: 'Explain quantum computing in one sentence'"
"List the files in the stabilityai/stable-diffusion-xl-base-1.0 model"

Direct API Call

curl -X POST https://mcp.aerostack.dev/s/aerostack/mcp-huggingface \
  -H 'Content-Type: application/json' \
  -H 'X-Mcp-Secret-HUGGINGFACE-TOKEN: hf_xxxxxxxxxxxx' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"search_models","arguments":{"query":"text-generation","sort":"downloads","limit":10}}}'

Security Notes

• HF tokens are injected at the Aerostack gateway layer — never stored in the worker
• Inference API has rate limits based on your HF plan (free tier: limited, Pro: higher)
• Gated models require accepting terms on the model page before inference works
• Some models may not be available for serverless inference — check model page for availability

License

MIT