x402agent/SolanaOS / skills/pdf-to-markdown
pdf-to-markdown
VerifiedConvert PDF documents to clean structured Markdown for LLM context. Supports two modes: fast (PyMuPDF) and accurate (IBM Docling TableFormer AI). Features aggressive persistent caching, image extraction with metadata, table detection, and batch processing. Use when asked to convert PDFs, extract PDF content, parse documents, or prepare PDF data for AI/LLM consumption.
PDF to Markdown Converter
Extract entire PDF content as clean, structured Markdown optimized for LLM context windows.
Quick Start
# Install dependencies (fast mode)
uv pip install pymupdf pymupdf4llm
# Convert a PDF
python pdf_to_md.py document.pdf
# Accurate tables (IBM Docling AI)
uv pip install docling docling-core
python pdf_to_md.py document.pdf --docling
Two Extraction Modes
Fast Mode (default)
Uses PyMuPDF with text-based table detection. Good for most documents.
python pdf_to_md.py report.pdf
# Output: report.md (with images/ directory if PDF has images)
- Speed: ~0.05 sec/page
- Tables: text-strategy detection (handles borderless tables)
- Images: extracted at source resolution
- Best for: simple layouts, text-heavy docs, quick processing
Accurate Mode (Docling)
Uses IBM's TableFormer AI model for ~93.6% table extraction accuracy.
python pdf_to_md.py report.pdf --docling
# First run downloads AI models (~2-3 min one-time setup)
- Speed: ~1 sec/page
- Tables: AI-powered structure detection (complex/borderless)
- Images: configurable resolution (default 4x)
- Best for: financial reports, scientific papers, complex layouts
Persistent Caching
Extractions are cached in ~/.cache/pdf-to-markdown/. Cache is keyed by file content hash + extraction mode — the same PDF extracts once and reuses forever.
# View cache stats
python pdf_to_md.py --cache-stats
# Clear cache for one PDF
python pdf_to_md.py document.pdf --clear-cache
# Clear entire cache
python pdf_to_md.py --clear-all-cache
Cache invalidates automatically when:
- Source PDF content changes (hash-based detection)
- Extractor version bumps (logic improvements)
- You explicitly clear it
Output Format
---
source: document.pdf
total_pages: 42
extracted_at: 2026-03-23T22:00:00
images_dir: images
---
# Document Title
Content here...
| Column A | Column B |
|----------|----------|
| data | data |

**[Image: figure_0001.png (1920x1080, 245.3KB)]**
---
## Extracted Images
| # | File | Dimensions | Size |
|---|------|------------|------|
| 1 | figure_0001.png | 1920x1080 | 245.3KB |
Usage Patterns
Basic Conversion
python pdf_to_md.py input.pdf # → input.md
python pdf_to_md.py input.pdf output.md # custom output path
python pdf_to_md.py input.pdf --docling # accurate tables
python pdf_to_md.py input.pdf --no-progress # suppress progress
Programmatic Usage
from extractor import extract_pdf_fast, extract_pdf_docling, get_page_count
# Fast extraction
pages = get_page_count("report.pdf")
markdown = extract_pdf_fast("report.pdf", image_dir="./images")
# Accurate extraction with images
markdown, image_paths = extract_pdf_docling(
"report.pdf",
output_dir="./images",
images_scale=4.0,
show_progress=True,
)
With Caching (recommended)
from pdf_to_md import CacheManager, ExtractionConfig, ImageManager
cache = CacheManager()
config = ExtractionConfig(pdf_path="report.pdf", docling=False)
valid, cache_key = cache.is_valid(config)
if valid:
result = cache.load(cache_key)
markdown = result.markdown
else:
# Extract and cache
markdown = extract_pdf_fast("report.pdf")
cache.save(cache_key, ExtractionResult(
markdown=markdown, image_dir=None, total_pages=pages
), config)
Batch Processing
# Convert all PDFs in a directory
for f in *.pdf; do python pdf_to_md.py "$f"; done
# Parallel with xargs
ls *.pdf | xargs -P4 -I{} python pdf_to_md.py {}
Architecture
pdf_to_md.py — CLI entry point, cache orchestration, output formatting
extractor.py — Extraction backends (fast + docling)
~/.cache/pdf-to-markdown/
└── {hash16}/
├── metadata.json — source info, mtime, version
├── full_output.md — cached markdown
└── images/ — cached extracted images
Key Classes
| Class | Purpose |
|---|---|
ExtractionConfig | PDF path, mode (fast/docling), image scale |
ExtractionResult | Markdown output, image dir, page count, cache flag |
CacheManager | Content-hash keying, atomic writes, validation |
ImageManager | Temp dir tracking, image metadata, path rewriting |
Cache Key Generation
SHA256(file_size | SHA256(first_64KB + last_64KB) | mode)[:16]
This means:
- Same PDF content → same cache key (regardless of filename)
- Different extraction modes → different cache entries
- File renames don't invalidate cache
- Content changes always invalidate
Dependencies
| Package | Mode | Purpose |
|---|---|---|
pymupdf | Both | PDF parsing, image extraction |
pymupdf4llm | Fast | Markdown conversion with table detection |
docling | Accurate | IBM document AI pipeline |
docling-core | Accurate | Document model and export |
Install
# Fast mode only
uv pip install pymupdf pymupdf4llm
# Both modes
uv pip install pymupdf pymupdf4llm docling docling-core
Image Handling
Images are extracted to an images/ directory next to the output markdown file. References use relative paths for portability.
- Fast mode: extracts all embedded images via xref deduplication
- Docling mode: extracts figures at configurable resolution (default 4x)
- Cache preserves images and copies them to output location on cache hit
- CMYK images auto-convert to RGB
Tips
- Use
--doclingfor financial reports, academic papers, or any document with complex table layouts - Cache stats help monitor disk usage:
--cache-stats - For LLM context windows, the metadata header provides page count and source info
- Page breaks use
<!-- PAGE_BREAK -->sentinels (not-----) to avoid false splits - Extractor version (
3.3.0) is tracked in cache to auto-invalidate on upgrades
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |