Productivity and Research
Stop Uploading PDFs to Claude
Every PDF you drag into Claude burns up to 5,000 tokens per page. MarkItDown from Microsoft (140k+ GitHub stars) converts them to text first. Cut token use by 70%.
Every PDF you drag into Claude burns up to 5,000 tokens per page because Claude processes the raw image instead of the text. MarkItDown is a free Microsoft tool with 140,000 or more GitHub stars that converts PDFs, Word documents, PowerPoints, spreadsheets, and HTML pages into clean markdown text. Run it before uploading and cut your token use by 70%, with better output quality on top of that.
How to do it
- 1Install MarkItDown: `pip install markitdown` (requires Python 3.8 or later)
- 2Convert a single PDF: `markitdown document.pdf > document.md` and you now have clean, searchable text
- 3Open the `.md` file to verify the conversion. Check that tables, headings, and key content came through correctly
- 4Upload the markdown file into Claude instead of the original PDF. The text is cleaner and the token count is a fraction of the image version
- 5Batch convert a whole folder: `for f in *.pdf; do markitdown "$f" > "${f%.pdf}.md"; done`
- 6MarkItDown also handles: DOCX, PPTX, XLSX, HTML, images with embedded text (via OCR), audio files (auto-transcribes), and YouTube URLs
Why it works
The token saving is significant but the quality improvement is bigger. Claude reads markdown text accurately. It reads scanned PDFs with errors, missed sections, and layout confusion that compounds through the entire response. Same source document, meaningfully better output.
All guides are free, no sign-up required. Browse free resources