Research Tracker MCP Server

A Model Context Protocol (MCP) server that provides research inference utilities. This server extracts research metadata from paper URLs, repository links, or research names using web scraping and API integration.

Features

• Author inference from papers and repositories
• Cross-platform resource discovery (papers, code, models, datasets)
• Research metadata extraction (names, dates, licenses)
• URL classification and relationship mapping
• Comprehensive research ecosystem analysis
• Rate limiting to prevent API abuse
• Request caching with TTL for performance
• Error handling with typed exceptions
• Security validation for all URLs
• Retry logic with exponential backoff

Frontend

The project includes a modern web interface built with Flask and vanilla JavaScript:

• Clean Design: Minimalist black and white theme with soft green accents
• Real-time Discovery: Live logging of the discovery process with scrollable output
• Responsive Layout: Grid-based design that adapts to different screen sizes
• Interactive Elements: Example URL buttons for quick testing
• Progress Tracking: Visual progress indicators and status updates
• Resource Display: Organized grid showing discovered papers, code, models, datasets, and demo spaces

UI Components

• Input Section: URL input field with discover button
• Discovery Log: Real-time scrolling log of the discovery process
• Results Grid: Clean display of discovered resources
• Example URLs: Pre-configured test cases for demonstration
• Status Indicators: Progress bars and status messages

Available MCP Tools

All functions are optimized for MCP usage with clear type hints and docstrings:

• infer_authors - Extract author names from papers and repositories
• infer_paper_url - Find associated research paper URLs
• infer_code_repository - Discover code repository links
• infer_research_name - Extract research project names
• classify_research_url - Classify URL types (paper/code/model/etc.)
• infer_publication_date - Extract publication dates
• infer_model - Find associated HuggingFace models
• infer_dataset - Find associated HuggingFace datasets
• infer_space - Find associated HuggingFace spaces
• infer_license - Extract license information
• find_research_relationships - Comprehensive research ecosystem analysis

Input Support

• arXiv paper URLs (https://arxiv.org/abs/...)
• HuggingFace paper URLs (https://huggingface.co/papers/...)
• GitHub repository URLs (https://github.com/...)
• HuggingFace model/dataset/space URLs
• Research paper titles and project names
• Project page URLs (github.io)

MCP Best Practices Implementation

This server follows official MCP best practices:

1. Security: URL validation, domain allowlisting, input sanitization
2. Performance: Request caching, rate limiting, connection pooling
3. Reliability: Retry logic, graceful error handling, timeout management
4. Documentation: Comprehensive docstrings with examples for all tools
5. Error Handling: Typed exceptions for different failure scenarios

Environment Variables

• HF_TOKEN - Hugging Face API token (required)
• GITHUB_AUTH - GitHub API token (optional, enables enhanced GitHub integration)

Usage

The server automatically launches as an MCP server when run. All inference functions are exposed as MCP tools for integration with Claude and other AI assistants.

Example

Test with the 3D Arena paper:

Input: https://arxiv.org/abs/2506.18787
Finds: dataset (dylanebert/iso3d), space (dylanebert/LGM-tiny), and more

Rate Limits

• 30 requests per minute per tool
• Automatic caching reduces duplicate requests
• Graceful error messages when limits exceeded

Error Handling

The server provides clear error messages:

• ValidationError: Invalid or malicious URLs
• ExternalAPIError: External service failures
• MCPError: Rate limiting or other MCP issues

Installation

1. Clone the repository
2. Install dependencies: pip install -r requirements.txt
3. Set environment variables
4. Run: python app.py

Requirements

• Python 3.8+
• See requirements.txt for dependencies

Running the Application

MCP Server Only

python app.py

Web Interface

python flask_app.py

The web interface will be available at http://localhost:5000

Gradio Interface (Alternative)

python ui.py

Project Structure

• app.py - Main MCP server entry point
• flask_app.py - Flask web interface
• ui.py - Gradio alternative interface
• mcp_tools.py - MCP tool implementations
• inference.py - Core inference logic
• discovery.py - Multi-round discovery functions
• static/ - CSS and JavaScript files
• templates/ - HTML templates
• utils.py - Utility functions