Skip to main content
Version: v1.6.47-dev1 🚧

Hugging Face MCP Server

Overview​

Hugging Face provides a large ecosystem of models, datasets, and AI tools. The Hugging Face MCP server enables Enterprise h2oGPTe agents to look up models and datasets and run inference via Hugging Face services.

For more information, visit Hugging Face.

When to use​

Use this MCP server when you want agents to:

  • Retrieve information about Hugging Face models and datasets
  • Run model inference via Hugging Face APIs
  • Search for models or datasets that match a given task

Requirements​

Before configuring this MCP server, ensure you have:

  • A Hugging Face account
  • A valid Hugging Face API token (HF_TOKEN)

Key capabilities​

The Hugging Face MCP server can:

  • Fetch metadata about models and datasets
  • Run inference requests against supported endpoints
  • Help agents discover appropriate models and datasets for a task

Configuration​

Add the following configuration in Agents > Tools > New Tool > Remote MCP Tools:

{
  "huggingface": {
    "command": "npx",
    "args": [
      "mcp-remote",
      "https://huggingface.co/mcp",
      "--header",
      "Authorization: Bearer os.environ/HF_TOKEN"
    ],
    "env": {
      "HF_TOKEN": "os.environ/HF_TOKEN"
    }
  }
}

Authentication​

In Agents > Authentication, create a secret with:

  • HF_TOKEN: Hugging Face API token with appropriate permissions

Notes​

  • This MCP server uses HTTP transport via the mcp-remote client.
  • Ensure that your Hugging Face token follows your organization’s security and rotation policies.
Feedback