MinerU — Extract LLM-Ready Data from Any Document

What it is

MinerU is an open-source document extraction tool that converts PDFs, scanned images, and complex documents into clean Markdown or structured JSON. It handles tables, formulas, images, and multi-column layouts that break simpler parsers. The output is designed to feed directly into RAG pipelines and LLM applications.

MinerU targets AI engineers building RAG systems, data scientists who need to process research papers at scale, and teams that ingest large document corpora for LLM training or retrieval.

How it saves time or tokens

Raw PDF text extraction often produces garbled output: broken table structures, missing formula text, and mangled multi-column layouts. MinerU's layout-aware parsing preserves document structure, which means your LLM receives clean, well-formatted context instead of noisy text. This reduces wasted tokens on malformed input and improves retrieval accuracy in RAG pipelines.

For batch processing, MinerU handles entire document directories in one command, eliminating manual per-file conversion workflows.

How to use

1. Install MinerU via pip: pip install magic-pdf[full]. Download the required model weights as documented in the README.
2. Run extraction on a PDF: magic-pdf -p input.pdf -o output/ -m auto. The -m auto flag automatically selects the best extraction mode for your document.
3. Find the output in the output/ directory as Markdown files with extracted images in a companion folder. Use the structured JSON output for programmatic pipelines.

Example

# Extract a research paper to Markdown
magic-pdf -p research-paper.pdf -o ./output -m auto

# Output structure:
# output/
#   research-paper/
#     auto/
#       research-paper.md      # Clean Markdown
#       images/                 # Extracted figures
#       research-paper.json     # Structured content

The Markdown output preserves headings, tables as Markdown tables, LaTeX formulas, and image references with extracted figure files.

Related on TokRepo

• AI tools for RAG — Tools for building retrieval-augmented generation pipelines
• AI tools for documents — Document processing and extraction solutions

Common pitfalls

• Model weights are large (several GB). The initial download takes time and disk space. Ensure you have sufficient storage before installing.
• Scanned PDFs with poor image quality produce lower-accuracy extractions. MinerU works best with high-resolution scans or born-digital PDFs.
• GPU acceleration significantly speeds up processing. CPU-only mode works but is much slower for large document batches.