Summary
PDF to Markdown that survives two columns and equations.
The value is in what it drops and what it keeps: page furniture goes, LaTeX and table structure stay, and the text comes back in reading order rather than column order. The trade is where the work happens — documents are sent to MinerU under your key, which decides whether this suits the papers you are converting.
What it is
A server that sends PDFs to the MinerU API and returns Markdown. Two tools: one for files on disk, one for URLs, both able to take several documents at a time.
What you get
- convert_pdf_file for local PDFs, convert_pdf_url for links
- Batch conversion, so a folder of documents is one call rather than twenty
- Headings, paragraphs and lists preserved as Markdown structure
- Formulas recognised and converted to LaTeX
- Tables extracted into structured form
- Reading order reconstructed for single-column, multi-column and complex layouts
- Headers, footers, footnotes and page numbers removed so the text stays coherent
Requirements
A recent Python, installed from a clone with uv. A .env file carries your MinerU API key and the addresses of the MinerU endpoints it calls. Conversion happens at MinerU, so the documents leave your machine.