Drop a PDF and get clean, LLM-ready Markdown: real tables, formulas as LaTeX, and OCR, with reading order intact. The same engine is a hosted REST API and an MCP endpoint, so AI agents and RAG pipelines convert PDFs as a built-in tool. It runs its own engines (MinerU + Docling), not an LLM wrapper. Free to start in the browser; also a Chrome extension and a ChatGPT Custom GPT.
No reviews yetBe the first to leave a review for PDF to Markdown
Maker
๐
Hey Product Hunt ๐
I'm Dmitry, an indie maker with a background in large-scale systems. I build small, focused tools that do one thing well.
PDFs are everywhere, but they're painful the moment you want to feed one to an LLM or a RAG pipeline: copy-paste drops the structure, mangles tables, and turns formulas into garbage. The good tools are either heavy self-hosted libraries or pricey per-page APIs.
So I built pdf2md.dev: PDF in, clean LLM-ready Markdown out. Real Markdown tables (not screenshots), formulas kept as LaTeX, OCR for scanned pages, reading order intact.
What makes it more than "another converter":
๐ A hosted REST API and a hosted MCP endpoint, so your agents can convert PDFs as a built-in tool
โ๏ธ It runs its own engines (open-source MinerU + Docling), not an LLM wrapper
๐งฉ Also a free web app (no sign-up), a Chrome extension and a ChatGPT Custom GPT, all on one engine
It's free to start in the browser, and files auto-delete (never used to train models).
I'd love your honest feedback, especially from folks doing RAG or agent workflows: what breaks, what's missing, what would make it your default? Try it on a gnarly PDF and tell me how it does ๐
Thanks for checking it out!
Report
Thanks for sharing! Quick question: how well does it handle complex PDFs with mixed tables, formulas, and multi-column layouts?
Honestly, that's the part we obsessed over. Tables go through a dedicated model, so merged cells and weird headers stay as real tables, not mush. Formulas come out as LaTeX, not screenshots. And two-column papers get read in the right order, not straight across the page.
I'll be real with you though. The nightmare case is a dense page with a formula stuck inside a table cell. That's hard for everyone. So we ship two engines and you can switch with one click. If one chokes on a doc, the other usually nails it.
Got a gnarly PDF lying around? Throw it at us. I'd love to see how it holds up.
Report
Congrats ! Is it OCR based? does it handle nestings (sub titles ..)?
Report
Maker
@racine_gย So, OCR: yes, but only when it's actually needed. If the PDF already has a text layer, we just read it. Clean and fast. OCR only kicks in on scanned pages. There's also a force-OCR toggle for those PDFs with a broken text layer. Works for English and Russian out of the box.
And yes, nesting is handled. Sub-titles keep their levels, so H1, H2, H3 all map to proper Markdown. Nested lists keep their indentation too. The whole outline survives instead of going flat.
@khlebobulย Thanks Gleb, appreciate it ๐ If you've got a gnarly PDF lying around, give it a spin and tell me where it trips up, that's the feedback I'm really after.
Thanks for sharing! Quick question: how well does it handle complex PDFs with mixed tables, formulas, and multi-column layouts?
@anton_iusย Thanks, that means a lot! ๐
Honestly, that's the part we obsessed over. Tables go through a dedicated model, so merged cells and weird headers stay as real tables, not mush. Formulas come out as LaTeX, not screenshots. And two-column papers get read in the right order, not straight across the page.
I'll be real with you though. The nightmare case is a dense page with a formula stuck inside a table cell. That's hard for everyone. So we ship two engines and you can switch with one click. If one chokes on a doc, the other usually nails it.
Got a gnarly PDF lying around? Throw it at us. I'd love to see how it holds up.
Congrats ! Is it OCR based? does it handle nestings (sub titles ..)?
@racine_gย So, OCR: yes, but only when it's actually needed. If the PDF already has a text layer, we just read it. Clean and fast. OCR only kicks in on scanned pages. There's also a force-OCR toggle for those PDFs with a broken text layer. Works for English and Russian out of the box.
And yes, nesting is handled. Sub-titles keep their levels, so H1, H2, H3 all map to proper Markdown. Nested lists keep their indentation too. The whole outline survives instead of going flat.
Really appreciate you digging in. ๐ฅ
@dmitry_petrakovย just tried it, it works !
Speed Reader
Amazing tool!
@khlebobulย Thanks Gleb, appreciate it ๐ If you've got a gnarly PDF lying around, give it a spin and tell me where it trips up, that's the feedback I'm really after.