MinerU
Local document parser converts PDFs, Office files and images into Markdown, JSON and LaTeX on macOS, Windows and Linux, with optional remote parsing.
MinerU parses documents locally into structured text for AI agents, RAG systems and knowledge bases. It's for people working with scanned PDFs, academic papers and Office files whose tables, formulas or page layouts need more care than plain text extraction.
It separates text and images while preserving document context and layout hierarchy. Alongside PDF, Word, PowerPoint and Excel, it accepts images, EPUB, OpenDocument files and web formats, including HTML and MHTML archives. The online service also accepts web URLs. Outputs include Markdown and JSON, with LaTeX and MathML for mathematical expressions.
Complex tables are a particular focus: it handles rotated tables, cells that span pages and merged cells, then exports table data as CSV, HTML or Markdown. Formula parsing covers long expressions, multiple lines and nested structures. EPUB parsing follows the book's reading order and preserves internal links.
Documents stay local by default. Remote parsing is optional and sends documents to a remote service; local failures don't silently trigger an upload. MinerU runs on macOS, Windows and Linux, and can serve parsing requests through a local server. CPU-only parsing works, with an ONNX path requiring as little as 2 GB RAM. Apple Silicon and NVIDIA GPUs can accelerate model-based parsing, with heavier GPU paths requiring more memory.
Supported engines include llama.cpp, PyTorch, vLLM, lmdeploy and MLX. MCP and OpenClaw support connect it to agent workflows, and it can export results to Dify and Notion.
