mineru-skill

A Claude Code skill for parsing documents with the MinerU API

Convert PDF, DOC, DOCX, PPT, PPTX, and images into clean Markdown/JSON
with OCR, formula recognition, table extraction, and batch processing.

Features

Quick Start

1. Install the Skill

cd ~/.claude/skills
git clone https://github.com/LeoLin990405/mineru-skill.git mineru

2. Set Up API Token

Get a free token at mineru.net/apiManage/token, then:

mkdir -p ~/.config/mineru
echo "YOUR_TOKEN" > ~/.config/mineru/token
chmod 600 ~/.config/mineru/token

3. Use It

In Claude Code — just ask naturally:

Parse this PDF to markdown: https://arxiv.org/pdf/2301.00001.pdf

Extract tables from report.pdf using the vlm model with OCR

Via CLI script:

# Parse from URL
./scripts/mineru-parse.sh https://example.com/paper.pdf --output ./parsed --extract

# Parse local file with VLM model
./scripts/mineru-parse.sh report.pdf --model vlm --ocr --output ./out

# Extra output formats
./scripts/mineru-parse.sh slides.pptx --format docx --format html

# Turn a book into an LLM-ready skill extraction workspace
./scripts/mineru-book-to-skill.sh book.pdf --title "My Book" --output ./workspaces --cloud-ok

CLI Reference

mineru-parse.sh <url_or_file> [options]

Options

Environment Variables

Models

API Limits (Cloud)

Book-to-Skill Workflow

First iteration support is implemented as a staging workflow. It prepares a
parsed book for a large language model, but it does not call an LLM and does
not install or enable generated skills automatically.

# Local files are uploaded to the MinerU cloud API; --cloud-ok is required.
./scripts/mineru-book-to-skill.sh ~/Books/example.pdf \
  --title "Example Book" \
  --output ./book-workspaces \
  --cloud-ok

The wrapper creates:

book-workspaces/
└── books/
    └── example-book/
        ├── README.md
        ├── source/
        ├── mineru/
        │   ├── parse_manifest.json
        │   ├── *_result.zip
        │   └── <extracted markdown files>
        └── analysis/
            └── book-skill-pack/
                ├── README.md
                ├── manifest.json
                ├── LLM_EXTRACTION_PROMPT.md
                ├── BOOK_SKILL_INDEX.md
                ├── MANAGE_SKILLS.md
                ├── source-markdown/
                └── skills/

Give LLM_EXTRACTION_PROMPT.md and source-markdown/ to a model. The model
should fill BOOK_SKILL_INDEX.md and create one candidate file per extracted
skill under skills/.

Expected model output is book-scoped:

• what the book can help the agent do
• when the agent should reference the book
• candidate workflows, checklists, diagnostics, decision rules, and prompt patterns
• source anchors and confidence
• which candidates should stay book-scoped vs be promoted to managed skills

Privacy Boundary

This repository uses the MinerU cloud API by default. Local book files are
uploaded to MinerU during parsing. Use a local MinerU workflow for private or
sensitive books. Generated packs do not store API tokens, authorization headers,
or remote result URLs.

Manage Skills Boundary

Book packs are candidate workspaces. Review the generated skills/*.md files
before promoting anything into a managed skill repository. Use your local skill
manager only after review, for example:

skills enable <promoted-skill-name>

Roadmap Notes

Next iterations should make extraction more structural:

• split the parsed book by chapters or section headings
• extract skill candidates from each chapter first
• synthesize the whole-book summary after chapter-level extraction
• keep generated skills packaged by the book's structure, so an agent can refer
  to the book as a coherent source package

The same pipeline should later generalize beyond books. Future inputs can
include video courses, papers, and other long-form text. Before extraction, the
system should classify the text type, such as book, course transcript, paper,
manual, or article collection, then choose the right segmentation and skill
extraction strategy.

Examples

See the examples/ directory for:

• parse_single.sh — Parse a single PDF from URL
• parse_local.sh — Upload and parse a local file
• parse_batch.py — Batch parse multiple documents (Python)
• book_to_skill.sh — Parse a book and stage an LLM-ready skill pack

Project Structure

mineru-skill/
├── SKILL.md                 # Claude Code skill definition (full API reference)
├── scripts/
│   ├── mineru-parse.sh      # CLI helper script
│   ├── mineru-book-to-skill.sh # Book parsing + skill-pack staging wrapper
│   └── book-skill-pack.sh   # Build a skill extraction pack from Markdown
├── examples/
│   ├── parse_single.sh      # Single URL parsing example
│   ├── parse_local.sh       # Local file parsing example
│   ├── parse_batch.py       # Batch processing example (Python)
│   └── book_to_skill.sh     # Book-to-skill workspace example
├── .github/
│   ├── ISSUE_TEMPLATE/      # Bug report & feature request templates
│   └── PULL_REQUEST_TEMPLATE.md
├── CONTRIBUTING.md
├── CHANGELOG.md
├── LICENSE                  # MIT
└── README.md

Documentation

Full API reference including all endpoints, request/response formats, error codes, and Python/curl examples is in SKILL.md.

Contributing

Contributions are welcome! Please read the Contributing Guide before submitting a PR.

Related Projects

• MinerU — The document parsing engine by OpenDataLab
• Claude Code — Anthropic's CLI for Claude

License

MIT © 2026 LeoLin990405