qiming/qiming-mcp-proxy/document-parser/README.md

Document Parser

English | 简体中文

---

Document Parser

A high-performance multi-format document parsing service supporting PDF, Word, Excel, and PowerPoint with GPU acceleration capabilities.

Features

• 🚀 High-Performance Parsing: MinerU and MarkItDown dual-engine support
• 🎯 GPU Acceleration: CUDA/sglang support for GPU acceleration (optional)
• 🔧 Zero-Configuration Deployment: Automatic environment detection and dependency installation
• 📚 Multi-Format Support: PDF, Word, Excel, PowerPoint, Markdown, and more
• 🌐 HTTP API: RESTful API interface for easy integration
• 📊 Real-time Monitoring: Built-in performance monitoring and health checks
• ☁️ OSS Integration: Alibaba Cloud OSS support for cloud storage

Quick Start

1. Environment Initialization

cd document-parser

# Initialize uv virtual environment and dependencies (first time)
document-parser uv-init

# Check environment status
document-parser check

2. Start Service

# Start document parsing service
document-parser server

# Or specify custom port
document-parser server --port 8088

The service will start at http://localhost:8087 (default) and automatically activate the virtual environment.

System Requirements

Basic Requirements

• Rust: 1.70+
• Python: 3.8+
• uv: Python package manager

GPU Acceleration (Optional)

• NVIDIA GPU: CUDA-compatible
• CUDA Toolkit: 11.8+
• GPU Memory: At least 8GB recommended

Supported Formats

Format	Parsing Engine	Features
PDF	MinerU	Professional PDF parsing, image extraction, table recognition
Word	MarkItDown	Document structure preservation, format conversion
Excel	MarkItDown	Table data extraction, format preservation
PowerPoint	MarkItDown	Slide content extraction, image saving
Markdown	Built-in	Real-time parsing, table of contents generation

Configuration

Basic Configuration

# Server configuration
server:
  port: 8087
  host: "0.0.0.0"

# MinerU configuration
mineru:
  backend: "vlm-sglang-engine"  # Enable GPU acceleration
  max_concurrent: 3
  quality_level: "Balanced"

GPU Acceleration Configuration

mineru:
  backend: "vlm-sglang-engine"  # Use sglang backend
  max_concurrent: 2              # Lower concurrency for GPU
  batch_size: 1

Common Commands

# Environment management
document-parser check              # Check environment status
document-parser uv-init            # Initialize environment
document-parser troubleshoot       # Troubleshooting guide

# Service management
document-parser server             # Start service
document-parser server --port 8088 # Specify port

# File parsing (CLI)
document-parser parse --input file.pdf --output result.md --parser mineru

API Usage

Parse Document

curl -X POST "http://localhost:8087/api/v1/documents/parse" \
  -H "Content-Type: multipart/form-data" \
  -F "file=@document.pdf" \
  -F "format=pdf"

Get Parsing Status

curl "http://localhost:8087/api/v1/documents/{task_id}/status"

API Documentation

Once the service is running, visit:

• OpenAPI Swagger UI: http://localhost:8087/swagger-ui/
• OpenAPI JSON: http://localhost:8087/api-docs/openapi.json

Performance Optimization

GPU Acceleration

1. Ensure sglang[all] is installed
2. Configure backend: "vlm-sglang-engine"
3. Adjust concurrency parameters based on GPU memory
4. Monitor GPU usage

Concurrency Control

mineru:
  max_concurrent: 2    # Adjust based on system performance
  batch_size: 1        # Process in small batches
  queue_size: 100      # Queue buffer size

Troubleshooting

Common Issues

1. Virtual environment not activated: Run source ./venv/bin/activate
2. Dependency installation failed: Run document-parser uv-init
3. GPU acceleration not working: Refer to CUDA Environment Setup Guide
4. Permission issues: Check directory and user permissions

Get Help

# Detailed troubleshooting guide
document-parser troubleshoot

# Environment status check
document-parser check

# View logs
tail -f logs/log.$(date +%Y-%m-%d)

Development

Build

cargo build --release

Test

cargo test

Code Check

cargo fmt
cargo clippy

License

This project is licensed under MIT License.

Contributing

Issues and Pull Requests are welcome!