Connect Hugging Face to Andromity with the Hugging Face MCP server

Hugging Face runs an official MCP server at https://huggingface.co/mcp. With it, Andromity can search the Hub for models, datasets, Spaces and papers, look up Hugging Face documentation, and call community tools from MCP-compatible Gradio Spaces while it writes your ML code.

Sign-in is through your Hugging Face account. Repository creation, Jobs, dynamic Space and sandbox tools require authentication; requests without a token get an anonymous tool set.

What the agent can do with Hugging Face

• Search and explore Hub models, datasets, Spaces and papers.
• Search Hugging Face documentation with natural language queries.
• Create repositories and write files to them (the Contribute Repos tool).
• Run, monitor and schedule Jobs on Hugging Face infrastructure.
• Create and use sandboxes, including sandbox file management.
• Call MCP-compatible Gradio Spaces as tools.

Example prompts

Ask for the work in plain language. Naming the server helps, for example “use the Hugging Face MCP server”.

• Find a small instruction-tuned model on Hugging Face that fits in 8 GB of memory, then update config/model.yaml and the loader in src/llm.py to use it.
• Search Hugging Face documentation for how to load LoRA adapters with PEFT, then fix the adapter loading in train.py.
• Find datasets on Hugging Face for weather time series and write a loader script that reads the most downloaded one.
• Find papers about vision-language models and summarize which ones are relevant to the captioning module in this repo.

Connect Hugging Face

1. You need a Hugging Face account.
2. Optional: open https://huggingface.co/settings/mcp while logged in to choose extra built-in tools and add Spaces.
3. In VS Code, open Andromity Hub → MCP Servers → Discover. In the terminal app, open /settings → MCP → + Connect an app. Find Hugging Face and select Connect.
4. Approve the sign-in in your browser. Andromity saves Hugging Face to ~/.andromity/mcp.json and connects it.

See Connect an app for how Andromity stores connectors and tokens.

Access and safety

• The agent acts as your Hugging Face account for the tools that need authentication, including repository creation and Jobs.
• You configure extra tools and Spaces at https://huggingface.co/settings/mcp.
• Hugging Face supports query parameters on the URL, such as ?no_image_content=true to drop image blocks from Gradio responses. To use one, add the server by hand in mcp.json.
• Andromity's permission modes apply to every tool call, including those that create repositories or start Jobs.

In an untrusted folder, MCP servers don’t start at all. Read workspace trust and permission modes for the full model.

Limits to know

• Hugging Face described the server as experimental when it announced it, and says experimental tools are added from time to time.
• Hugging Face's docs do not publish a full list of built-in tool names, so the capabilities above are by tool group.

Questions

Does Hugging Face have an official MCP server?

Yes. Hugging Face announced an official MCP server and hosts it at https://huggingface.co/mcp.

Can I add Spaces as tools?

Yes. Add MCP-compatible Gradio Spaces in your Hugging Face MCP settings, then reconnect or refresh the client.

Do I need a token to use it?

Not for the anonymous tool set. Repository creation, Jobs, dynamic Space and sandbox tools require Hugging Face authentication.

Sources

This guide is based on Hugging Face’s own documentation. The app controls its server, so tools can change; check the source for the current list.

• Hugging Face MCP Server docs
• Hugging Face MCP server on GitHub
• Hugging Face MCP server announcement

Hugging Face is a trademark of its owner. Andromity is not affiliated with or endorsed by Hugging Face.