mineru-doc-parser-skill

Imported from Yuling-Tian/mineru-doc-parser-skill (AGENTS.md). Install upstream with npx skills add Yuling-Tian/mineru-doc-parser-skill. Copyright stays with the author.

MinerU Document Parser

This repository contains a document parsing skill/tool that converts PDF, Word, PPT, and image files into clean Markdown using the MinerU cloud OCR API.

Key Files

• mineru-doc-parser/scripts/parse_doc.py — Main parser script (Python CLI)
• mineru-doc-parser/SKILL.md — Skill definition with full workflow instructions

How to Use

When the user wants to parse a document (PDF/DOCX/PPTX/PNG/JPG) to Markdown, run the parser script:

python mineru-doc-parser/scripts/parse_doc.py "<file_path>" --lang en

Workflow

1. Detect page count (PDF only)

pip install PyMuPDF -q 2>/dev/null
python -c "import fitz; d=fitz.open('<path>'); print(d.page_count); d.close()"

2. Choose API tier

• ≤ 20 pages → Agent API (no token needed)
• > 20 pages → Ask user: truncate first 20 pages or full parse?

3. Large PDFs (>20 pages)

For full parse, check for a persisted token:

python mineru-doc-parser/scripts/parse_doc.py --check-token

• Exit 0 → token exists, proceed
• Exit 1 → ask user for token (get one free at https://mineru.net/apiManage)

4. Run parser

python mineru-doc-parser/scripts/parse_doc.py "<file>" \
    --lang <en|ch|korean|japan|latin> \
    [--page-range "1-20"] \
    [--token "<token>"] \
    [--save-token] \
    [--ocr] [--no-table] [--no-formula]

5. Report results

Tell the user: output path, character count, API tier used, whether token was persisted.

API Reference

Config

• Token persistence: ~/.mineru/config.json (or ~/.claude/mineru_config.json for Claude Code users)
• Env var override: MINERU_API_TOKEN

Requirements

• Python ≥ 3.8
• requests (pip install requests)
• PyMuPDF (pip install PyMuPDF, optional — for PDF page count detection)