Hugging Face MCP server
Machine learning model pages on Hugging Face. Add one address to your AI assistant and it can call the tool below itself, with your FoxAPIs key. It costs the same credits as the REST API, and an answer with no data is not charged.
Server address
https://foxapis.com/mcp/data/hugging-face
Streamable HTTP, bearer key in the Authorization header. This address carries only the Hugging Face tool; https://foxapis.com/mcp/data carries all of them, and ?source=hugging-face,github picks several.
What you can ask
Five prompts to give your assistant once the server is added:
- How many downloads does bert-base-uncased have on Hugging Face?
- What licence is the Hugging Face model bert-base-uncased released under?
- Check when the Hugging Face model bert-base-uncased was last modified.
- Compare the likes and downloads of two Hugging Face models I name.
- What task is the model bert-base-uncased built for, and which library does it use?
The tool your assistant gets
| Tool | What it returns | Inputs | Credits a call |
|---|---|---|---|
huggingface_model | A Hugging Face model by id: downloads, likes, task, library, licence and last update. | model_id (required) | 2 |
Credits come from the same file as the pricing page. Plans and top-ups are on pricing.
Set it up
Use your FoxAPIs key (it starts with fxp_) as the bearer token. Get one free at foxapis.com/trial.
Claude Code
Run in a terminal. Its docs.
claude mcp add --transport http foxapis-data https://foxapis.com/mcp/data/hugging-face \ --header "Authorization: Bearer YOUR_KEY"
The Claude Code docs show this form (--transport http, name, URL, --header) for a server that takes a bearer token.
Cursor
.cursor/mcp.json in a project, or ~/.cursor/mcp.json for every project. Its docs.
{
"mcpServers": {
"foxapis-data": {
"url": "https://foxapis.com/mcp/data/hugging-face",
"headers": {
"Authorization": "Bearer YOUR_KEY"
}
}
}
}Cursor's remote-server example uses a url and a headers object.
VS Code (Copilot)
.mcp.json at the workspace root, or ~/.copilot/mcp-config.json. Its docs.
{
"mcpServers": {
"foxapis-data": {
"type": "http",
"url": "https://foxapis.com/mcp/data/hugging-face",
"headers": {
"Authorization": "Bearer YOUR_KEY"
},
"tools": [
"*"
]
}
}
}Do not store a real key in a file you share.
Cline
cline_mcp_settings.json (open it from Cline's MCP Servers panel). Its docs.
{
"mcpServers": {
"foxapis-data": {
"type": "streamableHttp",
"url": "https://foxapis.com/mcp/data/hugging-face",
"headers": {
"Authorization": "Bearer YOUR_KEY"
},
"disabled": false,
"autoApprove": []
}
}
}Cline's docs say that leaving type out selects the older SSE transport, so keep type.
Zed
settings.json (Zed: open settings). Its docs.
{
"context_servers": {
"foxapis-data": {
"url": "https://foxapis.com/mcp/data/hugging-face",
"headers": {
"Authorization": "Bearer YOUR_KEY"
}
}
}
}
n8n
an MCP Client Tool node in a workflow. Its docs.
Endpoint: https://foxapis.com/mcp/data/hugging-face Server transport: HTTP Streamable Authentication: Header Auth Name: Authorization Value: Bearer YOUR_KEY
Not run by us: the fields are the ones n8n's MCP Client Tool documents. The server speaks Streamable HTTP.
Claude Desktop and ChatGPT
Not covered here: their custom-connector screens ask for sign-in (OAuth) or no authentication, and this server takes a bearer key in a header, so it works best in the clients above. We have not tested it in either.
A real answer
The huggingface_model tool called once through a real MCP client session on 20 November 2026 with the input {"model_id":"bert-base-uncased"}. Long text is cut and lists are cut to three items for display; it is not invented.
{
"platform": "huggingface",
"model_id": "google-bert/bert-base-uncased",
"author": "google-bert",
"pipeline_tag": "fill-mask",
"library_name": "transformers",
"downloads_last_month": 33456723,
"likes": 3084,
"is_private": false,
"gated": false,
"license": "apache-2.0",
"languages": [
"tf",
"en"
],
"tags": [
"transformers",
"pytorch",
"jax"
],
"created_at": "2022-03-02T23:29:04.000Z",
"last_modified": "2024-02-19T11:06:12.000Z",
"siblings_count": 16,
"model_url": "https://huggingface.co/bert-base-uncased",
"paper_url": "https://arxiv.org/abs/1810.04805",
"credits_used": 2,
"latency_ms": 192
}
Charged: 2 credits.
Try it free
Runs the huggingface_model call with the sample input below. No key and no sign-up; a few runs a day per visitor.
{
"model_id": "bert-base-uncased"
}
Other data-source MCP servers
PubMed · bioRxiv · ClinicalTrials.gov · FDA · Steam · Apple App Store · all of them
Free to start: 100 credits, no card. Then from $10. Credits you buy never expire.
Get my free keyThe same data over REST: Hugging Face model. More: all MCP servers · people and company tools