MarkPrep
Runs 100% in your browser · nothing uploaded

Turn any document into AI-ready Markdown — with the token count, the chunks, and no upload.

Every converter says it's built for AI. Then it hands you a file and walks away. MarkPrep tells you how many tokens you're holding, whether it fits your model, and exactly where it breaks into chunks — all inside this tab, where the file never leaves.

Drop files, paste, or click to browse
PDF, Word, Excel, PowerPoint, HTML, CSV, JSON, EPUB, images and more. Nothing is uploaded.
Network requests that left this page during conversion: 0We're not asking you to trust a privacy policy. Press F12 → Network and watch.

Markdown is not the finish line. Your model is.

Every other converter in this category is marketed “for AI.” Not one tells you how many tokens you're holding. MarkPrep counts as it converts — pick your model and the number turns green or red.

47,300
tokens · exact
Fits GPT-4o (128k)

A live count, per model, updating as pages process.

312,000
tokens
Won't fit — here's the chunk plan

Red means it overflows. You immediately see how it breaks up.

#1Introduction › Scope812 tok
#2Methods › Sampling796 tok
#3Results › Table 3840 tok

Visible boundaries with heading paths, exportable as JSONL.

No upload
The engine ships to your browser and runs there.
No account
Convert unlimited files without signing up.
Works offline
Install once, convert on a plane.
Open formats
Standard Markdown and JSONL out.

Convert anything to Markdown

Honest comparison

Including where GPU libraries beat us. We publish the gaps instead of hiding them.

CapabilityMarkPrepHosted MarkItDownGPU tools
Runs in your browser (no upload)
Live token count & context verdict
Heading-aware chunk preview + JSONL
Works offline
Free & unlimitedLimited
Best-in-class scanned-PDF accuracy
Complex table reconstructionGoodBasicBest

It counts before you paste

Live token totals for GPT, Claude and Gemini, updating as each page converts, with a clear verdict against your model's context window. You stop pasting into an LLM and hoping.

Chunks you can actually see

Heading-aware boundaries in the preview, adjustable size and overlap, heading path and page number in every chunk, one-click JSONL export. Retrieval returns whole sections, not fragments.

Nothing leaves your machine

PDF parsing, OCR, table extraction — all compiled to WebAssembly, all in your tab. Convert files you're contractually forbidden from uploading, on a locked-down laptop, without a ticket.

The clean-up nobody else bothers with

Repeated headers, footers, page numbers and hyphenated line-breaks stripped automatically. Your embeddings stop absorbing page furniture.

We show you what we got wrong

Every table with a mismatched column count and every low-confidence scan is flagged, with the issue called out. You catch the broken row before your RAG pipeline does.

Your assistant can use it directly

Run our MCP server locally and Claude Code converts documents itself, on your machine. The conversion step disappears from your workflow.

$4 a month. Less than everyone.

Every conversion, every format, the token counts and the chunking — free, forever. Pro is $4 for the workflow around it: the local MCP server, watch folders, saved profiles and encrypted sync.

See pricing

Frequently asked questions

Why is this cheaper than everything else?+

Because conversions run on your computer instead of our servers, our fixed costs are about a dollar a month. We're not undercutting with venture money — we just have almost nothing to cover.

If it runs in my browser, is the quality worse?+

On Word, Excel, PowerPoint, HTML, CSV and digital PDFs, no — we run the same open-source parsers locally that hosted tools run remotely. For DOCX we use mammoth.js, the original implementation the Python version was ported from. On difficult scans a GPU tool beats us; our benchmark shows exactly where.

Are the token counts accurate?+

Exact for OpenAI models, which publish their tokenizer. Approximate for Claude and Gemini, which don't — we label those clearly rather than pretending otherwise.

What about large files?+

The limit is your browser's memory, not our policy. A 500-page PDF converts fine on a modern laptop. On a phone, expect slower results above 50 pages.

Do you see my files?+

No. There is no upload endpoint. Two exceptions are labelled every time: fetching a public URL, and bring-your-own-key OCR, which goes from your browser to your own provider using your own key.

Can I use this at work?+

That's who it's for. Nothing leaves the device, so most acceptable-use policies are satisfied without an exception request.

Why pay if conversion is free?+

You shouldn't, unless you want the local MCP server, watch folders, saved profiles or encrypted sync. Most people never need to.

Does it work offline?+

Yes. Install it as a PWA and it keeps converting with the Wi-Fi off — on a plane or an air-gapped machine.

Which formats are supported?+

PDF (digital and scanned), Word, Excel, PowerPoint, HTML, CSV/TSV, JSON, XML, EPUB, Jupyter notebooks, images, plain text, and public URLs.

Do I need an account?+

No. Conversion needs no account. You only sign up for Pro.