MinerU PDF Paper Parser

Problem

When reading PDF papers, copied text often loses formulas, figure captions, and layout, making it hard to feed content into deeper analysis. This skill focuses on paper-level PDF-to-Markdown conversion rather than plain OCR, targeting close reading, extracting text and images, and exporting to additional formats.

How It Works

It calls the MinerU parsing pipeline to turn PDF page structure into structured content, then generates full.md. The default model is vlm; you can also select pipeline or MinerU-HTML, and optionally export docx, html, or latex via --extra-formats. The skill uses uv run to manage dependencies, creating an isolated environment on first run and reusing .venv/ afterward.

Key steps include:
- Check daily quota before parsing, with a default limit of 2000 pages
- Run PDF parsing and write results to an output directory
- Keep optional intermediates such as layout.json, content_list.json, and model.json
- Record usage to /data/usage.log after completion

Boundaries And Notes

This skill depends on a stable network and the MinerU API, and a single run can take several minutes. Oversized files, non-standard PDFs, or network timeouts may affect results; quota resets daily based on local time, so estimate total pages before batch processing. Improper key configuration can block parsing, with precedence of CLI parameter, environment variable, then config file.