MinerU is a document parsing tool that converts PDF, image, DOCX, PPTX, and XLSX inputs into Markdown and JSON for retrieval, extraction, and downstream LLM processing. It originated in the pre-training data pipeline of InternLM, with a focus on symbol conversion in scientific literature, and it outputs text in human reading order with headers, footers, footnotes, and page numbers removed.
Three inference backends cover different trade-offs. The VLM model is MinerU2.5-Pro (1.2B), which handles image and chart parsing, truncated-paragraph merging, cross-page table merging, and image recognition inside tables; the pipeline backend uses PP-OCRv6, and the hybrid backend exposes an effort setting (medium by default, high for image analysis).
| Backend | Characteristics |
|---|---|
| pipeline | Model pipeline without generation, runs on CPU or GPU |
| vlm-engine | Vision-language model inference through vLLM, LMDeploy, or mlx |
| hybrid-engine | VLM layout plus native text extraction |
MinerU runs on Windows, Linux, and macOS in a pure CPU environment or with GPU or MPS acceleration, and supports Ascend, Cambricon, Enflame, MetaX, Moore Threads, Kunlunxin, Iluvatar, Hygon, Biren, and T-Head chips. It ships as a CLI, an HTTP API server, and a Gradio web UI, with Python, Go, and TypeScript SDKs, a Docker image, and an MCP server for Cursor, Claude Desktop, and Windsurf. It integrates natively with LangChain, LlamaIndex, RAGFlow, RAG-Anything, Flowise, Dify, and FastGPT, and a hosted web app and desktop client are available at mineru.net.
Features
- Input formats: PDF, images, DOCX, PPTX, and XLSX, plus web pages
- Layout reconstruction: reading-order output for single-column, multi-column, and complex layouts, preserving headings, paragraphs, and lists
- Formulas and tables: formulas converted to LaTeX and tables to HTML
- OCR: automatic detection of scanned or garbled PDFs, with recognition across 109 languages and handwriting support
- Images: extraction of images, image descriptions, table titles, and footnotes
- Output formats: multimodal and NLP Markdown, JSON sorted by reading order, and intermediate formats
- Visualization: layout and span visualizations for checking output quality
- Deployment surfaces: CLI, HTTP API server, Gradio web UI, Docker, and SDKs in Python, Go, and TypeScript
- MCP server: document parsing exposed to coding assistants
- Model sources: automatic model-source selection and local cache reuse on first install
Integrates with
Dify
Open-source LLM app development platform combining a visual workflow canvas, RAG pipelines, agents, model management, and observability
Flowise
Low-code visual builder for AI agents and LLM workflows, self-hostable or via managed cloud
LangChain
Framework for building LLM applications and agents from interoperable components and a large integrations library
LlamaIndex
Open-source data framework for building LLM and agentic apps over your own data, with 300+ integrations
