Stay free if you only need basic features. Upgrade if you need advanced features. Most solo builders can start free.
Yes. Docling is released under the Apache 2.0 license and the associated models (Docling layout, TableFormer, Granite-Docling, SmolDocling) are openly available on Hugging Face, so it can be embedded in commercial products and run on-premises without per-document fees.
Docling parses PDF, DOCX, PPTX, XLSX, HTML, Markdown, AsciiDoc, CSV, and images (PNG, JPEG, TIFF), and recent versions add audio transcription. Outputs include Markdown, HTML, JSON, and the structured DoclingDocument schema.
Docling runs locally with no data ever leaving your environment, which hosted APIs cannot offer. It also preserves richer structural information (tables via TableFormer, reading order, formulas) than most generic OCR APIs. The trade-off is that you operate the infrastructure yourself rather than paying per page.
Yes. Docling ships a Model Context Protocol (MCP) server so MCP-compatible agents and IDE assistants (Claude Desktop, Cursor, etc.) can call it as a tool to convert and chunk documents on demand, in addition to direct integrations with LangChain, LlamaIndex, Haystack, and Crew AI.
Yes. It integrates with OCR engines including EasyOCR, Tesseract, and RapidOCR, and can run vision-language pipelines (SmolDocling, Granite-Docling) that read directly from page images to produce structured output.
Start with the free plan — upgrade when you need more.
Get Started Free →Still not sure? Read our full verdict →
Last verified March 2026