Source: Docling Documentation
Official Docling project docs: supported input formats, export formats (Markdown, HTML, DocTags, JSON), OCR engines, VLM pipeline with GraniteDocling, chunkers, MCP server and framework integrations.
Contents
- What: official project documentation site
- Author/publisher: Docling project (originated at IBM Research)
- Date: living document; read 2026-11-14
- Original: docling-project.github.io/docling
Summary
The landing page of Docling’s docs lists what the library parses, what it exports, and how it plugs into LLM frameworks. It positions Docling as a local, self-hostable converter that produces a unified DoclingDocument representation, usable for RAG ingestion or agent tools.
Key claims
- Inputs (as of 2026-11): PDF, DOCX, PPTX, XLSX, HTML, EPUB, images (PNG, TIFF, JPEG), LaTeX, plain text, email, audio (via ASR), and recent additions such as OpenDocument and XBRL financial reports.
- Outputs: Markdown, HTML, DocTags, WebVTT and lossless JSON.
- PDF understanding: page layout, reading order, table structure, code, formulas, image classification; chart understanding can convert bar, pie and line charts into tables.
- OCR engines: pluggable, including Tesseract, RapidOCR and others, for scanned PDFs and images.
- VLM pipeline: an alternative pipeline that runs a vision-language model (GraniteDocling) over pages.
- Chunkers: hybrid (structure- and token-aware), line-based token, and trivial chunkers.
- Integrations: LangChain, LlamaIndex, CrewAI, Haystack and others; also a CLI, an API server (
docling-serve) and an MCP server; runs locally for air-gapped data.