MinerU

Active
GitHub Python NOASSERTION

Description

Transforms complex documents like PDFs into LLM-ready markdown/JSON for Agentic workflows, supporting layout analysis, formula recognition, and table extraction.

Key Features

  • Multi-format document parsing - Native support for PDF, DOCX, PPTX, XLSX, and images with Markdown/JSON output
  • VLM + OCR dual engine - Vision-language model combined with OCR, 109-language recognition, 95+ accuracy on OmniDocBench
  • Formula and table extraction - Auto-convert formulas to LaTeX, tables to HTML, with cross-page table merging and chart parsing
  • Multiple inference backends - pipeline (CPU-ready), vlm-engine, and hybrid-engine for flexible deployment
  • Rich integration ecosystem - MCP Server for Cursor/Claude, native integration with LangChain, Dify, FastGPT and other RAG frameworks
  • Enterprise-grade deployment - Docker, multi-GPU, mineru-router load balancing, compatible with domestic AI chips

Use Cases

💡 Document preprocessing for RAG systems, converting PDF/Word into structured data consumable by LLMs
💡 Scientific paper parsing with automatic extraction of formulas, tables, and references into editable formats
💡 Enterprise document batch conversion with automated pipelines for hundreds of pages
💡 Real-time document parsing and retrieval in AI coding tools via MCP Server integration

Tags

data-processing rag python llm tools

Categories

📚 RAG Tools

Quick Start

# Install MinerU
pip install uv
uv pip install -U "mineru[all]"

# Parse a document (GPU accelerated)
mineru -p input.pdf -o output/

# Pure CPU mode
mineru -p input.pdf -o output/ -b pipeline
Visit GitHub

Related Projects