MinerU

MinerU is a smart document parsing platform designed for LLMs, agents, and RAG systems. It takes various document formats and converts them into structured outputs like Markdown, JSON, and LaTeX. The platform preserves layout context and handles complex elements such as tables and mathematical formulas to assist AI applications and knowledge base construction.

• Parses PDF, Word, PPT, Excel, images, and web URLs
• Converts documents into Markdown, JSON, and LaTeX structured outputs
• Extracts and restores tables into CSV, HTML, or Markdown formats
• Transforms complex mathematical formulas into LaTeX and MathML standards
• Provides chemical paper analysis with molecular detection and chemical reaction extraction
• Supports MCP protocols and agent frameworks like Dify and Notion
• Offers an open source option with a GitHub repository

Pros

• Strong compatibility with multiple input formats ranging from standard office files to web URLs and images
• Detailed handling of specialized scientific content including mathematical formulas and chemical molecular structures
• Preserves layout context and document hierarchy to supply clean data for RAG systems and LLMs
• Provides workflow integrations and native support for Agent frameworks and tools

Cons

• Pricing details are not published on the page and require contacting the provider
• Requires checking specific API limits and deployment constraints directly with the platform
• User support and community resources rely primarily on external channels like GitHub and connected communities

MinerU fits developers, researchers, and technical teams building AI applications, retrieval-augmented generation systems, and autonomous agents that need clean, machine-readable document inputs. It is especially useful for handling scientific papers containing complex math, tables, and chemical structures. It is a poor fit for casual users who only need basic document viewing or simple text editing without programmatic parsing needs.

Unlike general file converters, MinerU is specifically tailored as a parsing engine for LLMs, agents, and RAG architectures. It goes beyond basic text extraction by preserving layout hierarchies, converting complex nested math into standard code, and parsing chemical structures for scientific workflows.

Dify, Notion, MCP protocol, OpenClaw, PDF, Word, PPT, Excel, images, web URLs, Markdown, JSON, LaTeX, CSV, HTML, and MathML