home / mcp / huggingface daily papers mcp server

HuggingFace Daily Papers MCP Server

A Model Context Protocol (MCP) server that provides read-only access to the Hugging Face Hub's papers.

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "huangxinping-huggingface-daily-paper-mcp": {
      "command": "uvx",
      "args": [
        "huggingface-daily-paper-mcp"
      ]
    }
  }
}

You can run the HuggingFace Daily Papers MCP Server locally and expose a programmable API for fetching HuggingFace daily papers. It supports date-based queries, today/yesterday lookups, paper metadata, and download links, all accessible through the MCP interface for easy integration with your tools and workflows.

How to use

You interact with the MCP server using a compatible MCP client. Start the server locally, then query for papers by date, or request today’s or yesterday’s papers. The MCP endpoints return structured paper data including title, authors, abstract, tags, votes, submission details, and links to the paper and PDF.

How to install

Prerequisites you need before starting: Python 3.10 or newer and the uvx runtime for easy MCP server execution.

Option A: Direct execution with uvx (recommended) start the server with this command.

uvx huggingface-daily-paper-mcp

Option 2: Local development

If you prefer to develop locally, clone the project and install dependencies, then run the server.

git clone https://github.com/huangxinping/huggingface-daily-paper-mcp.git
cd huggingface-daily-paper-mcp
uv sync

Local usage commands

Run as MCP Server (for development) to start the local server.

python main.py

Test the scraper function to verify data collection works as expected.

python scraper.py

Run the test suite to ensure code quality and MCP behavior.

uv run -m pytest test_mcp_server.py -v

Build the package if you need a distributable artifact.

uv build

MCP Interface

Tools are exposed through the MCP interface for retrieving papers by date, today’s papers, and yesterday’s papers.

Project structure and technical notes

The server is implemented in Python 3.10+ and uses standard MCP patterns. It relies on common libraries for HTTP requests and HTML parsing, with tests covering the MCP server behavior.

Example output

A single paper data entry includes title, authors, abstract, URLs, votes, and submission metadata. This lets you present or store paper information in a consistent format.

Configuration for MCP clients

Use an MCP client to connect to the local MCP server via the following standard runtime command.

{
  "mcpServers": {
    "huggingface_papers": {
      "command": "uvx",
      "args": ["huggingface-daily-paper-mcp"]
    }
  }
}

Security and maintenance notes

Ensure you manage access to your locally exposed MCP server and monitor logs for errors to maintain data quality and uptime.

Available tools

get_papers_by_date

Retrieve HuggingFace papers for a specific date using the date parameter formatted as YYYY-MM-DD.

get_today_papers

Retrieve today’s HuggingFace papers with no parameters.

get_yesterday_papers

Retrieve yesterday’s HuggingFace papers with no parameters.