PDF Parsing API — PDFs to Clean Markdown at Scale
A REST API that turns PDFs into structured Markdown: OCR for scanned pages, LaTeX formula recognition, table extraction across page breaks, and multi-column reading order — the document parsing layer for RAG pipelines, document AI products, and batch archives.
Quick start
Authenticate with a Bearer token or X-API-Key header and post a file:
curl -X POST 'https://www.kolmopdf.com/api/pdf-to-markdown-proxy/parse' \ -H 'Authorization: Bearer sk-xxx' \ -F 'file=@paper.pdf'
What the API handles
- PDF OCR API — scanned and photographed documents parsed the same as native PDFs.
- Table extraction — merged cells and tables that continue across pages come back as pipe tables.
- Math and code — formulas as LaTeX, code blocks fenced with language hints.
- Layout analysis — two- and three-column pages linearized into correct reading order.
- Inline translation — optionally translate to 8 languages during the same parse.
Built for RAG and document AI
Retrieval quality is capped by ingestion quality. Feeding raw PDF text extraction into a vector database scrambles reading order and destroys tables — the model retrieves noise. Parsing PDFs to Markdown first gives your chunker real heading boundaries, keeps tables machine-readable, and preserves equations as tokens an LLM can reason about. Teams use this pdf parser API as the first stage before embedding, for LLM fine-tuning corpora, and for migrating PDF archives into knowledge bases.
Frequently asked questions
What does the PDF parsing API return?
GitHub-Flavored Markdown with heading hierarchy, pipe tables, fenced code blocks, and LaTeX math in $ / $$ delimiters — ready for vector databases, LLM context windows, or content pipelines.
Does the API OCR scanned PDFs?
Yes. The parser is a Visual Language Model that reads pages as images, so scanned documents and native PDFs go through the same endpoint with no separate OCR configuration.
How is the API priced?
Per page parsed: 2 credits per page, or 3 credits per page with inline translation. New accounts include free credits, so you can integrate and test without a card.
Can it extract tables and math formulas?
Table extraction (including merged cells and tables spanning page breaks) and LaTeX formula recognition are the core of the model — it was trained on academic papers and technical documents where both are dense.
Is there a translation API as well?
Yes. A layout-preserving PDF translation API and a Markdown format conversion API share the same authentication. All three are covered in the API documentation.
Try it in the browser first: PDF to Markdown converter · Related: PDF translation, full API documentation.