---

name: docling description: | Document reading and conversion using Docling. Use this skill when user asks to read, open, or process document files in these formats: PDF, DOCX, PPTX, XLSX, HTML, Markdown, AsciiDoc, or images (PNG, JPG, TIFF). Supports OCR for scanned documents. Trigger when: (1) User asks to read/open a document file (e.g., "このPDFを読んで", "read this document", "ファイルの内容を確認して") (2) File extension is .pdf, .docx, .pptx, .xlsx, .html, .md, .adoc, .png, .jpg, .tiff (3) User wants to extract text from scanned documents with OCR (4) User wants to convert documents to Markdown/JSON/HTML (5) User wants to process documents with tables, figures, or photos (6) User wants to extract images/figures from documents

Docling Document Conversion

Convert documents (PDF, DOCX, PPTX, HTML, Markdown, etc.) to structured formats with image extraction.

Installation

Copypip install docling

# For page range processing (optional but recommended)
pip install pymupdf

# For Tesseract OCR (optional):
# macOS: brew install tesseract
# Ubuntu: apt-get install tesseract-ocr

Quick Start

CLI

Copy# Basic conversion (output to directory)
docling document.pdf --to markdown

# With OCR for scanned documents
docling scanned.pdf --ocr --ocr-engine easyocr --to markdown

# Batch conversion
docling file1.pdf file2.docx --output ./converted

Python

Copyfrom docling.document_converter import DocumentConverter

converter = DocumentConverter()
result = converter.convert("document.pdf")
print(result.document.export_to_markdown())

Using the Enhanced Conversion Script

Execute scripts/convert_document.py for advanced conversions with image extraction:

Copy# Basic PDF to Markdown with image extraction
python scripts/convert_document.py document.pdf -o ./output

# All formats (Markdown, JSON, HTML) with images
python scripts/convert_document.py document.pdf -o ./output -f all

# With OCR (Japanese + English)
python scripts/convert_document.py scanned.pdf -o ./output --ocr --languages ja en

# High accuracy table extraction
python scripts/convert_document.py document.pdf -o ./output --table-mode accurate

# Process specific page range
python scripts/convert_document.py large.pdf -o ./output --pages 1-20

# Generate batch script for large files (50+ pages)
python scripts/convert_document.py large.pdf -o ./output --generate-script

Output Structure

Copyoutput/
├── document.md       # Markdown with embedded image links
├── document.json     # Structured JSON data
├── document.html     # HTML output (optional)
└── images/
    ├── figure_001.png   # Diagrams, charts as PNG
    ├── figure_002.png
    ├── photo_001.jpg    # Photos as JPEG
    └── photo_002.jpg

Script Options

Large File Handling

For files with 50+ pages or 50+ MB:

Option 1: Page Range Processing

Copy# Process pages 1-20
python scripts/convert_document.py large.pdf -o ./output --pages 1-20

# Then process pages 21-40
python scripts/convert_document.py large.pdf -o ./output2 --pages 21-40

Option 2: Generate Batch Script

Copy# Generate a batch processing script
python scripts/convert_document.py large.pdf -o ./output --generate-script --batch-size 20

# Run the generated script
python ./output/batch_process.py

Advanced Configuration

OCR Setup

Copyfrom docling.document_converter import DocumentConverter, PdfFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import PdfPipelineOptions, EasyOcrOptions

pipeline = PdfPipelineOptions()
pipeline.do_ocr = True
pipeline.ocr_options = EasyOcrOptions(
    lang=["ja", "en"],  # Languages for OCR
    confidence_threshold=0.5
)

converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline)}
)

Table Extraction

Copyfrom docling.datamodel.pipeline_options import TableFormerMode

pipeline.do_table_structure = True
pipeline.table_structure_options.mode = TableFormerMode.ACCURATE  # or FAST
pipeline.table_structure_options.do_cell_matching = True

Image Extraction (Python API)

Copyfrom pathlib import Path
from docling.document_converter import DocumentConverter
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.datamodel.base_models import InputFormat
from docling.document_converter import PdfFormatOption

# Enable image generation
pipeline = PdfPipelineOptions()
pipeline.generate_picture_images = True

converter = DocumentConverter(
    format_options={InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline)}
)
result = converter.convert("document.pdf")

# Save images
images_dir = Path("./output/images")
images_dir.mkdir(parents=True, exist_ok=True)

for idx, picture in enumerate(result.document.pictures):
    if picture.image and picture.image.pil_image:
        pil_img = picture.image.pil_image
        pil_img.save(images_dir / f"image_{idx:03d}.png", "PNG")

Export Options

Copy# Markdown
markdown = doc.export_to_markdown()

# JSON (dict)
data = doc.export_to_dict()

# HTML
html = doc.export_to_html()

# Save with different image modes
from docling_core.types.doc import ImageRefMode
doc.save_as_markdown("output.md", image_mode=ImageRefMode.REFERENCED)

Supported Formats

OCR Engines

Common Patterns

Batch Processing

Copyfrom pathlib import Path

converter = DocumentConverter()
for pdf in Path("docs").glob("*.pdf"):
    result = converter.convert(str(pdf))
    output = pdf.with_suffix(".md")
    output.write_text(result.document.export_to_markdown())

RAG Chunking

Copyfrom docling.chunking import HierarchicalChunker

chunker = HierarchicalChunker()
chunks = list(chunker.chunk(result.document))
for chunk in chunks:
    print(chunk.text)

Extract Tables to CSV

Copyfor idx, table in enumerate(result.document.tables):
    df = table.export_to_dataframe()
    df.to_csv(f"table_{idx}.csv", index=False)
    print(df.to_markdown())  # Print as Markdown