Mineru Pdf

MinerU PDF

Overview

Parse a PDF locally with MinerU (CPU). Default output is Markdown + JSON. Use tables/images only when requested.

Quick start (single PDF)

# Run from the skill directory
./scripts/mineru_parse.sh /path/to/file.pdf

Optional examples:

./scripts/mineru_parse.sh /path/to/file.pdf --format json
./scripts/mineru_parse.sh /path/to/file.pdf --tables --images

When to read references

If flags differ from your wrapper or you need advanced defaults (backend/method/device/threads/format mapping), read:

• references/mineru-cli.md

Output conventions

• Output root defaults to ./mineru-output/.
• MinerU creates the per-document subfolder under the output root (e.g., ./mineru-output/<basename>/...).

Batching

Default is single-PDF parsing. Only implement batch folder parsing if explicitly requested.

MinerU PDF Parser - Clawdbot Skill

A Clawdbot skill for parsing PDFs locally using MinerU (CPU). Produces rich structured output including Markdown, JSON with layout data, and extracted images.

Features

• Local CPU processing — No GPU required; runs entirely on your machine
• Rich structured output — Markdown + detailed JSON with layout information
• Image extraction — Automatically extracts embedded images
• Table support — Optional table extraction (if supported by your MinerU version)
• Configurable — Flexible env overrides for different MinerU wrappers

Installation

Prerequisites

1. MinerU CLI installed and accessible (see MinerU installation) 2. Clawdbot installed

Install the skill

# Clone the repo
git clone https://github.com/kesslerio/MinerU-PDF-Parser-Clawdbot-Skill.git

# Or copy the mineru-pdf/ folder to your Clawdbot skills directory
cp -r MinerU-PDF-Parser-Clawdbot-Skill/mineru-pdf ~/.clawdbot/skills/

Usage

Quick start

# Run from the skill directory
./scripts/mineru_parse.sh /path/to/document.pdf

Options

./scripts/mineru_parse.sh /path/to/document.pdf --format json
./scripts/mineru_parse.sh /path/to/document.pdf --tables --images
./scripts/mineru_parse.sh /path/to/document.pdf --outroot ./my-output

Configuration

If your MinerU wrapper uses different flags, set env overrides. See mineru-pdf/references/mineru-cli.md for full documentation.

export MINERU_CMD=~/.local/bin/mineru
export MINERU_INPUT_FLAG=-p
export MINERU_OUTPUT_FLAG=-o

Output

MinerU creates a per-document subfolder under the output root:

./mineru-output/
└── document-name/
    └── auto/
        ├── document-name.md          # Markdown output
        ├── document-name_middle.json # Rich structured JSON (~50KB+)
        ├── document-name_layout.pdf  # Layout visualization
        └── images/                   # Extracted images

Output quality

MinerU produces rich structured output including:

• Layout-aware text extraction
• Detailed JSON with position/structure metadata
• Extracted images and layout PDFs

Best for: Documents requiring accurate layout preservation, image extraction, or structured data output.

Comparison with PyMuPDF

Use MinerU when: Quality and structure matter more than speed. Use PyMuPDF when: Speed matters or for simple text extraction.

License

Apache 2.0

Contributing

Issues and PRs welcome. Please test with a variety of PDFs before submitting changes.

Related

• PyMuPDF PDF Parser Skill — Fast, lightweight alternative
• MinerU — The underlying PDF parser
• Clawdbot — The AI agent framework

MinerU CLI Configuration

This skill assumes the MinerU wrapper described by Martin:

mineru -p <path_to_pdf> -o <output_dir> -b pipeline -l en -d cpu [-m auto]

Defaults (wrapper-based)

• Input flag: -p
• Output flag: -o
• Backend flag: -b
• Language flag: -l
• Device flag: -d (cpu)
• Method flag: -m (auto)
• Threads: OMP_NUM_THREADS=4 (set by wrapper or this script)

Output

MinerU will create a per-document directory under the output root. Typical outputs:

• <doc>.md
• <doc>_middle.json
• <doc>/ (images/tables)

Format mapping

The wrapper supports --format md|json|both and maps to MinerU values:

• md → markdown
• json → json
• both → markdown,json

If your wrapper uses different values, override:

export MINERU_FORMAT_FLAG=--format
export MINERU_FORMAT_VALUE_MD=markdown
export MINERU_FORMAT_VALUE_JSON=json
export MINERU_FORMAT_VALUE_BOTH=markdown,json

Advanced defaults

Defaults used by the wrapper:

• backend=pipeline
• method=auto
• device=cpu
• threads=OMP_NUM_THREADS (default 4)

Override flags/values via env vars as needed:

export MINERU_CMD=~/.local/bin/mineru
export MINERU_INPUT_FLAG=-p
export MINERU_OUTPUT_FLAG=-o
export MINERU_BACKEND_FLAG=-b
export MINERU_LANG_FLAG=-l
export MINERU_DEVICE_FLAG=-d
export MINERU_METHOD_FLAG=-m
export MINERU_THREADS_FLAG=OMP_NUM_THREADS

Test

./scripts/mineru_parse.sh /path/to/file.pdf --format both --threads 4 --lang en --backend pipeline --method auto --device cpu