tooldocumentspdfocrparsingmarkdownjsonragagentsUpdated 2026-06-27

MinerU

High-accuracy document parsing engine for turning PDFs, images, DOCX, PPTX, and XLSX files into LLM-ready Markdown/JSON.

What it is

MinerU is an OpenDataLab document parser focused on complex documents: PDFs, scanned pages, Office files, images, formulas, tables, and multi-column layouts. Its README says it can remove headers/footers/page numbers, output content in human reading order, extract images and tables, convert formulas to LaTeX, convert tables to HTML, detect scanned or garbled PDFs, and run OCR across 109 languages. Source: GitHub README, 2026-06-26

The GitHub repo describes MinerU as an agent/RAG document extraction tool with CLI, FastAPI, Gradio WebUI, SDK/API, Docker, MCP server, and framework integrations. The local route is mineru -p <input_path> -o <output_path>, with -b pipeline for CPU-friendly parsing when GPU acceleration is unavailable. Source: GitHub README, 2026-06-26

Routing

Use MinerU when the source is structurally difficult:

  • scanned PDFs or OCR-heavy documents
  • scientific papers with formulas
  • tables that need HTML or structured JSON
  • Office docs that should feed RAG or wiki ingest
  • large batches where reading order and visual layout matter

Keep MarkItDown - Universal File-to-Markdown Converter as the broad, lightweight default for ordinary files and Defuddle - Web Content Extraction for web pages. Escalate to MinerU when the output quality of lightweight conversion is the bottleneck.

Operational notes

Try a small sample first. MinerU's README explicitly warns that document parsing is difficult and recommends evaluating parsing quality before choosing deployment. That aligns with the wiki's test-before-bulk rule: run 3-5 documents, inspect Markdown/JSON, then batch. Source: GitHub README, 2026-06-26

For private or sensitive documents, prefer local execution over web demos. The project supports Windows, Linux, and macOS, with CPU, GPU/MPS, and OpenAI-compatible server modes; hardware requirements vary by backend. Source: GitHub README, 2026-06-26

Timeline

Related Pages

GitHub Starred Repos Harvest - 2026-06-26MarkItDown - Universal File-to-Markdown ConverterDefuddle - Web Content ExtractionRetrieval QualityRAG System ArchitectureSecurity and Review Skills

8 pages link here

Active Stack (What to Actually Use)MetaBrain CapsulesMetaCapability Routing MapMetaGitHub Starred Repos Harvest - 2026-06-26ResearchLangExtractToolsLiteParseToolsMarkItDown - Universal File-to-Markdown ConverterToolsMinerUTools