MinerU

MinerU turns complex documents into clean, machine-readable text for LLMs, RAG pipelines and AI agents. It comes from the OpenDataLab team and is one of the most-starred document-parsing projects on GitHub.

What it handles

• Inputs: PDF, Word, PowerPoint, Excel, images and web page URLs.
• Outputs: Markdown, JSON and LaTeX, keeping the reading order, headings and image-text context.
• Tables: rebuilds tables, including rotated tables, tables spanning pages and merged cells, as CSV, HTML or Markdown.
• Formulas: converts long and multi-line formulas to LaTeX or MathML.
• Chemistry: recognises molecular structure diagrams and reactions in scientific papers.

Ways to use it

• Web app and desktop client on mineru.net, with English and Chinese interfaces.
• API for batch and automated processing, with MCP support so agents can call it, and export to tools such as Dify or Notion.
• Self-hosted: run it yourself from the open-source code (a Python SDK, CLI and a Gradio web UI). The license is Apache 2.0 with added terms that require a separate commercial license only for very large organisations; check the repository before commercial use.

Good for

Anyone feeding papers, reports or scanned documents into an AI knowledge base, for example with Dify or Open WebUI.