microsoft/markitdown
Python tool for converting files and office documents to Markdown.
Convert any whitepaper to markdown, feed it to Claude, flag the red flags fast.
Turn what you learned into a concrete stack decision.
Want the shortlist in your inbox?
Subscribe for the weekly brief that turns new AI noise into the few tools and workflows worth testing.
Python tool for converting files and office documents to Markdown.
Guide
5 AI Repos Worth Your Time This Week (March 26)
markitdown finally solves the doc-to-LLM problem. Plus 4 more repos that earned a star.
Guide
The File Tool Every Crypto Researcher Is Sleeping On
Stop copy-pasting whitepapers into ChatGPT. There's a better way.
Guide
2 Weeks of AI Tool Trends: The Infrastructure Layer Is Winning
Agents are the headline. Infrastructure is where the real momentum is.
Most crypto projects die by PDF. They bury the tokenomics in a 40-page whitepaper, stack it with jargon, and count on you not doing the math. The good news: you don't have to do the math manually anymore.
Here's the exact workflow I use to run DD on a new project in under 5 minutes — no dev skills required.
PDFs are designed for humans, not AI. When you paste a whitepaper into Claude directly, you get a wall of poorly-formatted text, broken tables, and garbled charts. Claude tries its best but it's working with garbage input. Garbage in, garbage out.
The fix is a one-step conversion layer before you touch AI at all.
markitdown is a Microsoft open-source tool that converts PDFs, Word docs, and web pages into clean markdown. It's free, runs locally, and handles most document formats you'll actually encounter in crypto — whitepapers, tokenomics decks, litepages, even some Gitbook exports.
Install it once:
pip install markitdown
Then convert any doc:
markitdown whitepaper.pdf > whitepaper.md
That's it. You now have a clean, structured markdown file that Claude can actually work with. Tables stay intact. Sections have headers. The math is readable.
For web-based docs (like a Notion litepaper or a Gitbook), pass the URL directly:
markitdown https://docs.someproject.xyz > project.md
This is where most people go wrong. They paste the doc and ask "what do you think?" You get a generic summary that's basically the project's own marketing copy parroted back.
Don't do that. Be specific about what you want flagged.
Here's the prompt I actually use:
I'm doing DD on a crypto project. Read this whitepaper and flag: (1) vague or unverifiable claims with no mechanism explained, (2) tokenomics math that doesn't add up — supply percentages, vesting schedules, unlock timing, (3) missing info you'd expect in a legitimate project, (4) anything that sounds like it was written to sound impressive rather than be accurate. Be blunt.
Paste that, then paste the full markdown content. Claude will work through it section by section and give you an actual critical read — not a cheerleader summary.
In the last few months this workflow caught two projects worth walking away from before I'd spent any serious time on them.
Project A: The tokenomics said 40% to ecosystem fund, 30% to team (4-year vest), 20% to investors (1-year cliff), 10% to public. Adds up to 100% on paper. But the litepaper also mentioned a "community rewards program" drawing from a separate 15% allocation. Claude flagged it in 30 seconds. The numbers didn't reconcile. When I pushed the team on it, they couldn't explain it either.
Project B: Every claim in the "technology" section was described in business outcomes ("we enable seamless cross-chain liquidity") with zero mechanism. No technical spec, no architecture diagram, no link to a codebase. Claude flagged every single one as unverifiable. The project had been pitching to funds. No one had apparently asked the obvious question.
Neither of these required technical knowledge to spot — just a structured read with the right questions. That's what the prompt does.
markitdown isn't perfect. Heavily formatted PDFs with complex layouts sometimes come out messy — embedded images get lost, multi-column layouts occasionally jumble. For most standard whitepapers it's fine. For heavily designed pitch decks, you may get some noise in the output.
Claude also can't verify claims against external sources. It can flag that something sounds vague or that math is inconsistent within the document — it can't tell you whether the team actually has the experience they claim or whether the product is live on mainnet. You still have to do that verification manually or with other tools.
This workflow is a first-pass filter, not a replacement for deeper research. It just makes the first pass fast enough that you can run it on every project instead of only the ones that already passed your gut check.
| Step | Tool | Time | |------|------|------| | Convert whitepaper/litepaper to markdown | markitdown | ~30 seconds | | Feed to Claude with DD prompt | Claude (any tier) | ~2–3 minutes | | Review flagged issues | You | ~2 minutes |
Total: under 5 minutes for a structured critical read of any project doc.
If a project can't survive a 5-minute AI sanity check on their own published documents, that tells you everything you need to know.
Written by McKlaud AI. Want to know which AI tools actually fit your business? Get a free AI audit.