MinerU PDF Parser

Parse PDFs, Word docs, PPTs, and images into clean Markdown using MinerU's VLM engine. Use when: (1) Converting PDF/Word/PPT/image to Markdown, (2) Extractin...

Security Scan

The skill's code, instructions, and requested environment access are consistent with a document-to-Markdown parser that uploads files to mineru.net using a MINERU_TOKEN; no obvious mismatches or hidden exfiltration channels were found.

Purpose & Capability

Name/description (convert PDFs/Word/PPT/images to Markdown) aligns with the provided scripts and API usage. The scripts call mineru.net endpoints, accept files, and produce Markdown — functionality matches the stated purpose.

Instruction Scope

SKILL.md instructs the agent to set MINERU_TOKEN and run provided Python scripts, upload files, and save extracted output locally or to an Obsidian path. All referenced operations (reading input files, uploading to mineru.net, downloading/ extracting zip results, writing output) are within the expected scope. Minor inconsistency: scripts/mineru_async.py only enumerates PDF files (glob *.pdf) while SKILL.md claims async version supports all listed formats; this is likely an implementation detail but worth noting.

Install Mechanism

Dependencies are standard PyPI packages (requests, aiohttp) and there is a requirements.txt. No suspicious remote downloads or archive extraction from untrusted URLs are used. This is a typical pip-based installation footprint.

Credentials

The only required credential is MINERU_TOKEN (Bearer API token), which is proportionate for a cloud parsing service. One metadata glitch in the registry shows 'Required env vars: [object Object]' — a UI/metadata bug but not a functional request for additional unrelated secrets.

Persistence & Privilege

Skill is not always-enabled and does not request persistent elevated privileges or modify other skills. It runs only when invoked and only uses the provided token and local filesystem paths for input/output.

Assessment

This skill uploads your documents to mineru.net for parsing — do not use it with confidential or regulated documents unless you trust the service and its privacy policy. Provide only the MINERU_TOKEN (no other credentials required). Note the minor metadata bug in the registry (shows [object Object]) and that the async script currently enumerates only PDFs; if you need DOCX/PPTX images in async mode use the other scripts or confirm behavior. Installing the skill will require installing standard Python packages (requests, aiohttp). Verify output paths (especially cloud/Obsidian paths) before running to avoid accidental writes or syncs.

MinerU Document Parser

Convert PDF, Word, PPT, and images to clean Markdown using MinerU's VLM engine — LaTeX formulas, tables, and images all preserved.

Setup

1. Get free API token at https://mineru.net/user-center/api-token

export MINERU_TOKEN="your-token-here"

Limits: 2000 pages/day · 200 MB per file · 600 pages per file

Supported File Types

Type	Formats
📕 PDF	.pdf — papers, textbooks, scanned docs
📝 Word	.docx — reports, manuscripts
📊 PPT	.pptx — slides, presentations
🖼️ Image	.jpg, .jpeg, .png — OCR extraction

Commands

Single File

python3 scripts/mineru_v2.py --file ./document.pdf --output ./output/

Batch Directory with Resume

python3 scripts/mineru_v2.py \
  --dir ./docs/ \
  --output ./output/ \
  --workers 10 \
  --resume

Direct to Obsidian

python3 scripts/mineru_v2.py \
  --dir ./pdfs/ \
  --output "~/Library/Mobile Documents/com~apple~CloudDocs/Obsidian/VaultName/" \
  --resume

Chinese Documents

python3 scripts/mineru_v2.py --dir ./papers/ --output ./output/ --language ch

Complex Layouts (Slow but Most Accurate)

python3 scripts/mineru_v2.py --file ./paper.pdf --output ./output/ --model vlm

CLI Options

--dir PATH          Input directory (PDF/Word/PPT/images)
--file PATH         Single file
--output PATH       Output directory (default: ./output/)
--workers N         Concurrent workers (default: 5, max: 15)
--resume            Skip already processed files
--model MODEL       Model version: pipeline | vlm | MinerU-HTML (default: vlm)
--language LANG     Document language: auto | en | ch (default: auto)
--no-formula        Disable formula recognition
--no-table          Disable table extraction
--token TOKEN       API token (overrides MINERU_TOKEN env var)

Model Version Guide

Model	Speed	Accuracy	Best For
pipeline	⚡ Fast	High	Standard docs, most use cases
vlm	🐢 Slow	Highest	Complex layouts, multi-column, mixed text+figures
MinerU-HTML	⚡ Fast	High	Web-style output, HTML-ready content

Script Selection

Script	Use When
mineru_v2.py	Default — async parallel (up to 15 workers)
mineru_async.py	Fast network, need maximum throughput
mineru_stable.py	Unstable network — sequential, max retry

Output Structure

output/
├── document-name/
│   ├── document-name.md    # Main Markdown
│   ├── images/             # Extracted images
│   └── content.json        # Metadata

Performance

Workers	Speed
1 (sequential)	1.2 files/min
5	3.1 files/min
15	5.6 files/min

Error Handling

• 5x auto-retry with exponential backoff
• Use --resume to continue interrupted batches
• Failed files listed at end of run

API Reference

For detailed API documentation, see references/api_reference.md.