Does MinerU send my document anywhere?

Yes on the default path. It uses MinerU’s official Flash API, which needs no token, and the document is uploaded to mineru.net for processing, with a free tier of roughly 10 MB and 20 pages. For confidential documents consider a locally running parser such as MarkItDown or PaddleOCR instead.

Does MinerU work on both OpenClaw and Hermes?

Yes, from the same Tools page on either framework. On Hermes, restart the bot after installing if it does not notice the tool.

When should I use MinerU instead of MarkItDown?

Use MinerU for hard PDFs — scans with no text layer, equations, and multi-column layouts where reading order matters. MarkItDown covers a wider range of file types and runs locally, which makes it the better default for ordinary documents.

← Home

Guide

MinerU

Some PDFs are text in a wrapper. Others are a photograph of a page, a two-column layout with footnotes, or an equation. MinerU handles the second kind, converting them into Markdown your agent can actually reason over. Works on OpenClaw and Hermes.

The Problem It Solves

Ask a bot about a scanned contract without a parser and you get confident nonsense, because there is no text to read — only pixels. Multi-column academic papers are worse in a subtler way: naive extraction interleaves the columns, so the words are all present and the sentences are gibberish.

MinerU is built for exactly those documents: scans, formulas, and tables, out as structured Markdown with the reading order intact.

Install It

Open the dashboard and go to Tools.
Pick a running bot — tools install per instance.
Install MinerU. No account or token is needed to start.

On Hermes, restart the bot afterwards if it does not appear to notice the tool.

Know Where the Document Goes

This one is different from a purely local tool, and the difference is worth stating plainly before you feed it anything sensitive.

The default path is MinerU’s official Flash API, which needs no token and is why the tool works immediately. Your document is uploaded to mineru.net to be processed, and that free tier is bounded — around 10 MB and 20 pages per document.

For a public paper or a product manual that is a fine trade. For a signed contract, a medical record or anything under an NDA, decide deliberately rather than by default — and consider whether MarkItDown or PaddleOCR, which run in your own container, are enough for that document.

What to Ask For

“Read this scanned invoice and give me the line items.”
“Pull the results table out of this paper as Markdown.”
“Summarise this two-column report — the layout is confusing the usual reader.”
“Extract the equations from these lecture notes.”

You do not name the tool. Send the file, say what you want from it, and the agent reaches for the parser it has.

Which Document Tool for Which Job

MinerU — hard PDFs: scans, formulas, complex layouts. Uploads to the MinerU service on the default free path.
MarkItDown — a wide range of file types flattened to Markdown, locally.
Pandoc — conversion between formats, locally.
OfficeCLI — authoring and editing Office files, locally.

Troubleshooting

A large document fails. The free Flash path caps size and page count. Split the file, or send the pages that matter.
The output is still messy. Some scans are genuinely poor. Ask for a specific region — a table, a section — rather than the whole document.
The bot used a different parser. With several document tools installed the agent chooses; name the hard case (“this is a scan”) to push it toward MinerU.

Try It

Install it and send the worst PDF you have — a phone photo of a printed page is a good test. Ask for the text as Markdown and compare with what the bot managed before.