📄 Document Analyzer MCP Server

English | 简体中文

Make AI understand complex documents - MCP server solving AI context limitations

🎯 Key Features

• ✅ Smart Document Analysis - Auto-detect sections, handle merged cells

• ✅ Multi-format Support - Excel (.xlsx, .xls) | PDF/Word in development

• ✅ Precise Field Mapping - Field mapping table + section-level reading

• ✅ High Performance - Structured caching + lazy loading

🚀 Quick Start

Installation

macOS / Linux (Recommended with pipx)

# Install pipx
brew install pipx  # macOS
# or sudo apt install pipx  # Ubuntu/Debian

# Install doc-mcp-server
pipx install doc-mcp-server

Windows

pip install doc-mcp-server

For more installation options, see Full Installation Guide

Configure Claude Code

Add to ~/.claude.json or your project's config file:

{
  "mcpServers": {
    "document-analyzer": {
      "command": "doc-mcp-server"
    }
  }
}

For detailed configuration, see Quick Start Guide

📚 Full Documentation

• Installation Guide - Platform-specific installation steps

• Update Guide - How to upgrade to the latest version

• Quick Start - Configuration and basic usage

• Usage Guide - Complete API and examples

• Troubleshooting - Common issues and solutions

💡 Usage Example

# 1. Analyze document structure
analyze_document(file_path="/path/to/document.xlsx")

# 2. Read specific section
read_section(file_path="/path/to/document.xlsx", section_name="Section 1")

# 3. Read single field
read_field(file_path="/path/to/document.xlsx", field_key="Section1_CompanyName")

🛠️ Available Tools

🎯 Why Use This?

Problem: Large Excel files consume massive tokens when directly read by AI

• ❌ Traditional: Read entire 323-row Excel → 15000+ tokens → Often fails

• ✅ Using MCP: Structured reading → 2000 tokens → 90%+ success rate

Performance Improvements:

• 🚀 Token consumption reduced by 87% (15000 → 2000)

• ✅ Success rate improved from 30% to 90%+

• ⚡ Handles 323 rows × 24 columns with 4249 merged cells

🤝 Contributing & Feedback

• Report Issues: GitHub Issues

• Contribute Code: CONTRIBUTING.md

📄 License

MIT License - see LICENSE for details

Made with ❤️ by Yang Jiahui