QuantConnect PDF MCP Server

An advanced Model Context Protocol (MCP) server that provides intelligent search and retrieval capabilities for QuantConnect PDF documentation. This server converts PDFs to searchable markdown format and provides fast, context-aware search using TF-IDF scoring and proximity matching.

Features

• Intelligent PDF Processing: Automatically converts PDFs to structured markdown with proper formatting

• Fast Search Index: Uses inverted index with TF-IDF scoring for relevant results

• Context-Aware Results: Returns relevant excerpts with highlighted matches

• Caching System: Avoids reprocessing unchanged PDFs for better performance

• Proximity Matching: Boosts results where query terms appear close together

• Three MCP Tools: Search, list documents, and retrieve full content

Project Structure

Installation

Prerequisites

• Python 3.8 or higher

• pip package manager

Step 1: Install Dependencies

Install required packages:

The requirements.txt includes:

• mcp - Model Context Protocol library

• PyPDF2 - PDF text extraction

• asyncio - Asynchronous processing

Step 2: Prepare Your Environment

Create a virtual environment (recommended):

Configuration

Claude Desktop Setup

Find your Claude Desktop configuration file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• Linux: ~/.config/claude/claude_desktop_config.json

Add this configuration (adjust paths to match your system):

Environment Variables

• QUANTCONNECT_PDF_FOLDER: Directory containing your PDF files (required)

• QUANTCONNECT_MARKDOWN_FOLDER: Directory for converted markdown files (optional, defaults to PDF_FOLDER/markdown)

Usage

Starting the Server

1. Standalone testing:

   export QUANTCONNECT_PDF_FOLDER="/path/to/your/pdfs" python server.py

2. With Claude Desktop: Restart Claude Desktop after configuration to load the MCP server

3. Manual PDF conversion (optional):

   python convert_pdfs.py [pdf_folder] [markdown_folder]

Testing the Integration

Test in Claude by asking:

• "Can you list the available QuantConnect documents?"

• "Search for information about backtesting in the QuantConnect docs"

• "What does the QuantConnect documentation say about indicators?"

• "Show me page 5 of the Local Platform documentation"

Available MCP Tools

The server provides three powerful tools accessible through Claude:

1. search_quantconnect_docs

Purpose: Intelligent search through all QuantConnect documentation Parameters:

• query (required): Search terms or topic to find

• max_results (optional): Number of results to return (default: 5)

Features:

• TF-IDF scoring for relevance ranking

• Proximity matching for multi-word queries

• Context extraction with highlighted matches

• Returns document excerpts with page numbers

2. list_quantconnect_docs

Purpose: List all available PDF documents in the collection Parameters: None

Returns: Complete catalog of processed documents with metadata

3. get_document_content

Purpose: Retrieve full content from specific documents Parameters:

• filename (required): Document name (with or without .md extension)

• page_number (optional): Specific page to retrieve

Use cases: Reading complete sections, accessing specific pages, extracting code examples

Technical Architecture

Search Engine

• Inverted Index: Maps words to document locations for fast lookup

• TF-IDF Scoring: Balances term frequency with document rarity

• Proximity Boosting: Enhances results where query terms appear together

• Context Extraction: Provides relevant snippets around matches

Caching System

• PDF Processing Cache: Avoids reprocessing unchanged files using MD5 hashes

• Search Index Cache: Persists search index for faster startup

• Incremental Updates: Only processes new or modified PDFs

Performance Features

• Asynchronous Processing: Non-blocking PDF conversion and indexing

• Background Initialization: Server starts immediately while processing continues

• Efficient Storage: Markdown conversion reduces memory usage vs. raw PDF text

Troubleshooting

Common Issues

1. Server not connecting

   • Verify absolute paths in Claude Desktop configuration

   • Check Python virtual environment activation

   • Ensure server.py has execute permissions

2. PDFs not loading

   • Confirm QUANTCONNECT_PDF_FOLDER path exists

   • Check PDF file permissions and readability

   • Look for error messages in server output

3. Search returning no results

   • Wait for initial PDF processing to complete

   • Check if markdown files were created successfully

   • Try broader search terms

4. Performance issues

   • Ensure adequate disk space for markdown files

   • Check if antivirus is scanning the project folder

   • Consider moving cache files to faster storage

Debug Mode

Run the server with debug output:

Advanced Usage

Bulk PDF Processing

Process all PDFs without starting the server:

Custom Search Queries

The search supports various query types:

• Single terms: backtesting

• Multi-word queries: custom indicator development

• Technical terms: OnData event handler

• Code concepts: Algorithm.Initialize method

Integration Examples

Ask Claude sophisticated questions like:

Contributing

To extend the server:

1. Add new document formats: Extend the conversion system in server.py:236

2. Improve search: Enhance the SearchIndex class for semantic search

3. Add specialized tools: Create domain-specific search functions

4. Performance optimization: Implement parallel processing or database storage

Version History

• v0.3.0: Enhanced search with TF-IDF scoring and proximity matching

• v0.2.0: Added caching system and background processing

• v0.1.0: Basic PDF to markdown conversion and simple search