I built file2markdown.ai because pasting PDFs into LLMs produces garbage: broken tables, lost headings, silent truncation. It converts documents to clean, structured Markdown — and since I first wrote this post, it's become properly automatable, so this is a full update with corrected details.
What it handles: PDF, DOCX, PPTX, XLSX, CSV, JSON, XML, HTML, EPUB, images and scanned PDFs (OCR), audio, ZIP. PDFs go through pdf-inspector (Rust), Office/web formats through Microsoft's MarkItDown, scans through AI OCR.
Why not just ask your LLM for "the markdown"? Because the model retypes your document — paraphrased sentences, invented table cells, truncation on anything long. This runs real extraction engines: exact text, exact tables, same output every time.
The automation story (new): there's now an MCP server. Add https://mcp.file2markdown.ai/mcp to Claude, Cursor, or any MCP client and convert URLs and files mid-conversation — it's in the official MCP registry as ai.file2markdown/file2markdown. The site's pages also register WebMCP tools for browser agents (live in stock Chrome 149+ via origin trial). Setup: file2markdown.ai/mcp
Honest limits: 5 conversions/day with no account, 15/day with a free account, 25MB files. Pro removes the limits, adds 100MB files, batch, OCR, and API keys for the MCP server. Files are deleted immediately after conversion — nothing is stored.
If you've ever burned an afternoon fixing a mangled table paste, this is the fix
Top comments (0)