An AI Skill that transforms PDF documents into clean Markdown using MinerU's VLM engine.
中文文档 | English
A Skill is an AI capability package that extends your AI assistant's abilities. When you ask the AI to do something, it automatically:
- Recognizes the task from your natural language
- Activates the appropriate skill
- Executes the task using the skill's tools
- Delivers results back to you
You: 解析这些考研数学真题 PDF 到我的 Obsidian
AI: 📚 发现 40 个 PDF 文件
⏳ 开始并行解析 (5 workers)...
✅ 1993年考研数学（一）真题 → Markdown
✅ 1994年考研数学（一）真题 → Markdown
...
✅ 完成！已保存到 Obsidian/考研/数学一/
npx skills add Nebutra/MinerU-SkillSupported: OpenCode, Claude Code, Codex, Cursor, 35+ more
# Clone to your skills directory
git clone https://github.com/Nebutra/MinerU-Skill.git ~/openclaw-skills/mineru/
# Set API token
export MINERU_TOKEN="your-token-here" # Get from https://mineru.net/user-center/api-token# Install via clawhub CLI
clawhub install mineru# Clone to AI skills folder
git clone https://github.com/Nebutra/MinerU-Skill.git ~/.claude/skills/mineru/把 ./document.pdf 解析成 Markdown
解析 ./papers/ 目录下的所有 PDF，输出到 ./output/
把这些 PDF 解析后直接保存到我的 Obsidian Vault
解析 1987-2025 年考研数学真题，保存到 Obsidian/考研/数学一/
用 10 个并发，跳过已处理的文件
| Feature | Description |
|---|---|
| 📄 PDF Input | Local files, URLs, batch directories |
| 📝 Output | Markdown + JSON metadata + Images |
| 🔢 LaTeX | Math formulas preserved |
| 📊 Tables | Structure extraction |
| 🖼️ Images | Auto-extracted to images/ |
| ⚡ Async | 15x parallel uploads |
| 🔄 Resume | Skip processed files |
| 📁 Obsidian | Direct vault output |
You can also use directly via CLI:
# Single file
python scripts/mineru_v2.py --file ./doc.pdf --output ./output/
# Batch with resume
python scripts/mineru_v2.py \
--dir ./pdfs/ \
--output ~/Obsidian/MyVault/ \
--workers 10 \
--resume| Option | Description |
|---|---|
--dir PATH |
Input directory |
--file PATH |
Single file |
--output PATH |
Output directory |
--workers N |
Concurrency (default: 5) |
--resume |
Skip processed files |
--token TOKEN |
API token |
output/
├── document-name/
│ ├── document-name.md # Main Markdown
│ ├── images/ # Extracted images
│ │ ├── image_0_0.png
│ │ └── ...
│ └── content.json # Metadata
└── ...
Test: 10 PDFs, ~15 pages each (MacBook Air M1)
| Configuration | Time | Speed |
|---|---|---|
| Sequential | 8.5 min | 1.2 files/min |
| Async (5 workers) | 3.2 min | 3.1 files/min |
| Async (15 workers) | 1.8 min | 5.6 files/min |
- Visit MinerU
- Create free API token
- Set environment:
export MINERU_TOKEN="your-token-here"Free Tier: 2000 pages/day, 200MB max file
┌─────────────────────────────────────────────────────────────┐
│ USER REQUEST │
│ "Parse these PDFs to Markdown" │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ AI ASSISTANT │
│ • Recognizes PDF parsing task │
│ • Activates MinerU skill │
│ • Reads SKILL.md for instructions │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ MINERU SKILL ENGINE │
│ Scanner ──► Scheduler ──► Worker Pool (N workers) │
│ │ │
│ ▼ │
│ API: Get URL ──► Upload ──► Poll ──► Download │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ OUTPUT │
│ Markdown + JSON + Images ──► Obsidian/Files │
└─────────────────────────────────────────────────────────────┘
- Fork → Branch → Commit → Push → PR
MIT License - see LICENSE
If this skill helps you, give it a ⭐!
Made with ❤️ by Nebutra
