Mineru PDF Converter

MinerU PDF Converter

Convert PDF and other documents to high-quality Markdown using MinerU cloud API. Handles large PDFs (>600 pages) automatically by splitting, converting, and merging.

Capabilities

• Convert PDF, images, and other documents to Markdown/LaTeX
• Preserve formulas, tables, and complex layouts
• Auto-upload local files via MinerU batch API
• Auto-split large PDFs (>600 pages) and merge results
• Support additional output formats: LaTeX, DOCX, HTML

Requirements

Python packages:

pip install requests pymupdf

API Token: Token stored at: ~/.claude/skills/mineru-pdf-converter/references/mineru-token.md

Quick Start

Basic Conversion

python ~/.claude/skills/mineru-pdf-converter/scripts/mineru_convert.py \
  --input "/path/to/document.pdf" \
  --token-file "~/.claude/skills/mineru-pdf-converter/references/mineru-token.md"

Convert from URL

python ~/.claude/skills/mineru-pdf-converter/scripts/mineru_convert.py \
  --url "https://example.com/paper.pdf" \
  --token-file "~/.claude/skills/mineru-pdf-converter/references/mineru-token.md"

Verbose Mode with Progress

python ~/.claude/skills/mineru-pdf-converter/scripts/mineru_convert.py \
  --input "/path/to/document.pdf" \
  --token-file "~/.claude/skills/mineru-pdf-converter/references/mineru-token.md" \
  --verbose

Output shows progress percentage during conversion:

Uploading file: /path/to/document.pdf
File uploaded, batch_id: abc123
Waiting for conversion...
Status: running (25/100 pages, 25.0%)
Status: running (50/100 pages, 50.0%)
Status: done
Downloading result...

Additional Formats

# Include LaTeX output
python ~/.claude/skills/mineru-pdf-converter/scripts/mineru_convert.py \
  --input "/path/to/document.pdf" \
  --token-file "~/.claude/skills/mineru-pdf-converter/references/mineru-token.md" \
  --extra-formats "latex"

Workflow

When user requests PDF conversion:

1. Identify input type

   • Local file path: Use batch upload API to get temporary URL
   • URL: Submit directly to conversion API

2. Check file size (for PDFs)

   • If >600 pages: Split into 500-page chunks using PyMuPDF
   • Process each chunk separately
   • Merge final Markdown output

3. Execute conversion

   python ~/.claude/skills/mineru-pdf-converter/scripts/mineru_convert.py \
     --input "[path]" \
     --token-file "~/.claude/skills/mineru-pdf-converter/references/mineru-token.md"
   

4. Report result

   • Confirm output path (subfolder named after input file by default)
   • Note any warnings or partial failures
   • Provide path to main .md file

Parameters Reference

Page Ranges

The --page-ranges parameter allows you to convert only specific pages:

# Convert pages 1-50 and 100-150
python ~/.claude/skills/mineru-pdf-converter/scripts/mineru_convert.py \
  --input "/path/to/document.pdf" \
  --page-ranges "1-50,100-150" \
  --token-file "~/.claude/skills/mineru-pdf-converter/references/mineru-token.md"

How it works:

• Local files (--input): Pages are extracted client-side using PyMuPDF before upload
• URL input (--url): Page ranges sent to MinerU API server-side

This means page ranges now work for both local files and URLs.

Large PDF Handling

PDFs over 600 pages are automatically:

1. Split into chunks of max 500 pages using PyMuPDF
2. Each chunk converted separately via the API
3. Output Markdown files merged in order with page markers
4. Temporary chunk files cleaned up

To handle large PDFs, ensure PyMuPDF is installed:

pip install pymupdf

Model Selection

Error Handling

Output Structure

The conversion produces a ZIP file that is extracted to a subfolder named after the input file. This prevents naming conflicts when converting multiple PDFs in the same directory.

Default behavior (no --output-dir specified):

For input /path/to/paper.pdf, output is saved to /path/to/paper/:

/path/to/paper/
├── full.md               # Main Markdown file
├── images/               # Extracted images
│   ├── image_1.png
│   └── image_2.png
└── paper.json            # Structured content (optional)

With --output-dir specified:

When --output-dir /custom/path is provided, files are extracted directly to that directory (no subfolder created):

/custom/path/
├── full.md
├── images/
│   └── ...
└── paper.json

API Quota

• High priority: 2000 pages/day
• Low priority: Additional capacity (slower processing)
• Check remaining quota in the API response

Supporting Files

• scripts/mineru_convert.py - Main conversion orchestrator
• scripts/pdf_splitter.py - PDF splitting utility (PyMuPDF)
• scripts/merge_markdown.py - Output merger for chunked conversions
• references/api-reference.md - Full MinerU API documentation

Troubleshooting

Token Expired

The JWT token has an expiration date. If authentication fails:

1. Log in to mineru.net
2. Get a new API token
3. Update ~/.claude/skills/mineru-pdf-converter/references/mineru-token.md

Conversion Hangs

For very large or complex documents:

1. Increase timeout: --timeout 1200
2. Use page ranges to convert in sections: --page-ranges "1-100"
3. Try the pipeline model: --model pipeline

Missing Formulas or Tables

Ensure recognition is enabled (default):

• --enable-formula true
• --enable-table true

For detailed API reference, see references/api-reference.md.