mcpField
H
Huggingface Daily Papers

Huggingface Daily Papers

Local
github.com/huangxinpingAI & Agents7Updated 61w ago

About Huggingface Daily Papers

Huggingface Daily Papers is an MCP server for researchers and agents that need to browse HuggingFace papers by date. It returns paper details and links through both tools and resources.

Key features

  1. 01Fetch papers by date
  2. 02Return titles, authors, and abstracts
  3. 03Include tags, votes, and submitters
  4. 04Provide paper and PDF links
  5. 05Support tools and resources

Use cases

  1. 01Browse today's or yesterday's papers
  2. 02Find papers from a specific date
  3. 03Retrieve abstracts and PDF links

MCP support

The server exposes date-based paper lookup tools and resources for today's and yesterday's papers to MCP clients.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"huggingface-daily-paper": {
"command": "uvx",
"args": [
"huggingface-daily-paper-mcp"
]
}
}
}
Run commanduvx huggingface-daily-paper-mcp

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
pypihuggingface-daily-paper-mcp

Tools

3

Saved tool list · updated 2w ago

ParameterTypeRequiredDescription
datestringYesDate in YYYY-MM-DD format

Details

From the project README

HuggingFace Daily Papers MCP Server

A MCP (Model Context Protocol) server for fetching HuggingFace daily papers.

Features

  • Fetch today's, yesterday's or specific date HuggingFace papers
  • Provides paper title, authors, abstract, tags, votes, and submitted by info
  • Includes paper links and PDF download links
  • Supports MCP tools and resource interfaces
  • ArXiv integration for complete author lists
  • Complete error handling and logging
  • Comprehensive test coverage

Installation & Usage

Option 1: Direct execution with uvx (Recommended)

Install and run directly using uvx:

uvx huggingface-daily-paper-mcp

This will automatically install the package and its dependencies, then start the MCP server.

Option 2: Local development

For local development, clone the repository and install dependencies:

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):

python main.py

Test Scraper Function:

python scraper.py

Run Tests:

uv run -m pytest test_mcp_server.py -v

Build Package:

uv build

MCP Interface

Tools
  1. get_papers_by_date

    • Description: Get HuggingFace papers for a specific date
    • Parameters: date (YYYY-MM-DD format)
  2. get_today_papers

    • Description: Get today's HuggingFace papers
    • Parameters: None
  3. get_yesterday_papers

    • Description: Get yesterday's HuggingFace papers
    • Parameters: None
Resources
  1. papers://today

    • Today's papers JSON data
  2. papers://yesterday

    • Yesterday's papers JSON data

Project Structure

huggingface-daily-paper-mcp/
├── main.py                    # MCP server main program
├── scraper.py                 # HuggingFace papers scraper module
├── test_mcp_server.py         # MCP server test cases
├── README.md                  # Project documentation
├── .gitignore                 # Git ignore file
├── pyproject.toml             # Project configuration file
└── uv.lock                    # Dependency lock file

Tech Stack

  • Python 3.10+: Programming language
  • MCP: Model Context Protocol framework
  • Requests: HTTP request library
  • BeautifulSoup4: HTML parsing library
  • pytest: Testing framework
  • uv: Python package manager

Development Standards

  • Use uv native commands for package management
  • Follow Python PEP 8 coding standards
  • Include type hints and docstrings
  • Complete error handling and logging
  • Write unit tests to ensure code quality

Example Output

Single paper data structure:

{
  "title": "CMPhysBench: A Benchmark for Evaluating Large Language Models in Condensed Matter Physics",
  "authors": ["Weida Wang", "Dongchen Huang", "Jiatong Li", "..."],
  "abstract": "CMPhysBench evaluates LLMs in condensed matter physics using calculation problems...",
  "url": "https://huggingface.co/papers/2508.18124",
  "pdf_url": "https://arxiv.org/pdf/2508.18124.pdf",
  "votes": 15,
  "submitted_by": "researcher123",
  "scraped_at": "2025-08-27T10:30:00.123456"
}

MCP Tool output format:

Title: CMPhysBench: A Benchmark for Evaluating Large Language Models in Condensed Matter Physics
Authors: Weida Wang, Dongchen Huang, Jiatong Li, Tengchao Yang, Ziyang Zheng...
Abstract: CMPhysBench evaluates LLMs in condensed matter physics using calculation problems...
URL: https://huggingface.co/papers/2508.18124
PDF: https://arxiv.org/pdf/2508.18124.pdf
Votes: 15
Submitted by: researcher123
--------------------------------------------------

AI IDE/CLI Configuration

Claude Code (CLI)

Add to your MCP configuration:

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

Add to your .cursorrules or MCP settings:

{
  "mcp": {
    "servers": {
      "huggingface-papers": {
        "command": "uvx",
        "args": ["huggingface-daily-paper-mcp"],
        "env": {}
      }
    }
  }
}
Windsurf IDE

Add to your Windsurf MCP configuration:

{
  "mcpServers": {
    "huggingface-papers": {
      "command": "uvx",
      "args": ["huggingface-daily-paper-mcp"]
    }
  }
}
VS Code with Continue Extension

Add to your continue configuration:

{
  "mcp": {
    "servers": {
      "huggingface-papers": {
        "command": "uvx",
        "args": ["huggingface-daily-paper-mcp"]
      }
    }
  }
}
Other MCP-Compatible Tools

For any MCP-compatible client, use:

# Command
uvx huggingface-daily-paper-mcp

# Or with Python path
python -m main

License

MIT License

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.