A fast, independent resource for installing, configuring and using Docling for document parsing, OCR, Markdown, JSON and RAG workflows. No account, no backend - just the commands and answers you need.
Pick a task and jump straight to a command, tool or guide.
Copy-pasteable CLI commands for the most common document conversion tasks.
Convert a local PDF into clean, structured Markdown.
docling convert report.pdf --to mdDownload and convert an online document directly from an HTTP(S) URL.
docling convert https://arxiv.org/pdf/2408.09869 --to mdExport the full DoclingDocument JSON schema, including bounding boxes and reading order.
docling convert report.pdf --to jsonForce full-page OCR on image-only pages that have no selectable text.
docling convert scan.pdf --ocr-mode full_page --to mdExport HybridChunker chunks that preserve headings, tables and metadata for vector stores.
docling convert report.pdf --to chunks --chunks-type hybridExpose Docling to AI desktop clients (Claude Desktop, LM Studio) over the Model Context Protocol.
uvx --from=docling-mcp docling-mcp-serverWhat Docling can read and write, based on the official supported-formats list.
Exports: Markdown, JSON, YAML, HTML, split-page HTML, text, DocTags, DocLang, DocLang archive, WebVTT, LaTeX and chunks (JSONL).
Pick the OCR backend that matches your platform and accuracy needs.
Short, task-oriented recipes with CLI and Python for the most common jobs.
Turn documents into structure-aware chunks that keep headings, tables and page metadata.
from docling.document_converter import DocumentConverter
from docling.chunking import HybridChunker
converter = DocumentConverter()
result = converter.convert("report.pdf")
chunker = HybridChunker()
for chunk in chunker.chunk(result.document):
print(chunk.text)How Docling compares to other document conversion tools in the AI ingestion space.
Unlike legacy text extractors such as PyPDF or pdfplumber, Docling performs full visual layout analysis, reading-order recovery and table reconstruction with TableFormer - and it runs locally without forcing a cloud API.
| Capability | Docling | Unstructured | PyMuPDF / pdfplumber | Marker | MinerU |
|---|---|---|---|---|---|
| Document understandingWhether the tool reconstructs reading order, layout and tables — or just dumps text. | |||||
| Layout & reading order ? | Yes | Yes (hi_res) | Yes (PyMuPDF4LLM + Layout) | Yes | Yes |
| Table structure ? | Yes (TableFormer) | Yes (hi_res) | Yes (PyMuPDF4LLM) | Yes | Yes |
| Formulas & math ? | Optional enrichment | Not documented | Not documented | Yes (LaTeX) | Yes |
| Text extraction & OCRDigital PDFs need fast extraction; scans need OCR. No single engine wins both. | |||||
| Scanned-document OCR ? | Yes (pluggable) | Yes | Yes (automatic OCR) | Yes (surya VLM) | Yes |
| OCR engine choice ? | 7 engines | Tesseract / PaddleOCR | Tesseract or rapidocr (installed separately) | Built-in (surya VLM) | Built-in |
| Inputs & outputsWhat goes in (formats) and what comes out for LLMs (Markdown, JSON, chunks). | |||||
| Input formats ? | 36 formats (84 extensions) | dozens of file types (OSS) | PDF; Office via PyMuPDF Pro | PDF, images, Office, HTML, EPUB | PDF, images, Office, OpenDocument, EPUB, HTML |
| Markdown output ? | Yes | Via element conversion | Yes | Yes | Yes |
| Structured JSON ? | Yes (DoclingDocument) | Yes (elements) | Yes | Yes | Yes (content list) |
| Chunk-ready structure ? | Yes (HybridChunker) | Yes (chunk_by_title) | Page chunks | Yes (chunks output) | Locator-based reads |
| Deploy & operationsWhere it runs, what hardware it needs, and what it costs to operate. | |||||
| Local / offline execution ? | Yes (no telemetry) | Yes (telemetry on by default) | Yes | Yes | Yes (remote is opt-in) |
| Hardware needs ? | CPU ok, GPU faster | CPU ok | CPU only | CPU (fast) / GPU (balanced; NVIDIA needs Docker + NVIDIA Container Toolkit) | CPU (basic, ONNX; standard/advanced also run on CPU); GPU for PyTorch engines |
| Install weight ? | models download on first use | Varies by extras | ~25 MB core wheel | Model weights via surya | model size varies by tier |
| License ? | MIT | Apache-2.0 (library) | AGPL-3.0 or commercial (pdfplumber: MIT) | Apache-2.0 code; OpenRAIL-M weights | MinerU Open Source License |
| AI / RAG fitFramework connectors, agents and multimodal inputs for production RAG. | |||||
| RAG integrations ? | LangChain, LlamaIndex, Haystack and more | Yes (see docs) | LlamaIndex, LangChain | Not documented | Agent skill and SDK |
| Audio / video (ASR) ? | Yes (Whisper, optional asr extra) | Not documented | Not documented | Not documented | Not documented |
| Cloud API option ? | Self-host (docling-serve) | Yes (paid API) | No hosted API | Yes (Datalab platform) | Official service (opt-in) |
| Cost & trustPrice per page, governance and how to verify claims yourself. | |||||
| Self-host cost ? | Free (MIT) | Free OSS / paid API | Free (AGPL) / paid license | Free code; weights capped | Free under license terms |
| Governance ? | Linux Foundation AI & Data | Unstructured (company) | Artifex Software | Datalab | OpenDataLab |
| Best for ? | Local RAG over mixed PDFs | Multi-format ETL | Fast text extraction | Books & papers | Papers & Office mix |
Docling runs locally on Windows, macOS (Apple Silicon) and Linux (x86_64 and ARM64).
Choose options in your browser and get an executable command or Python snippet.
docling convert report.pdf \
--to md \
--ocr-engine rapidocr \
--ocr-mode default \
--table-mode accurate \
--device auto \
--num-threads 4Run Docling on your own documents in the browser - no installation required.
Quick fixes for the most frequent installation, GPU, OCR and conversion problems.
Everything you need to turn raw documents into clean, structured data for LLM context windows.
Parse PDF, DOCX, PPTX, XLSX, HTML, EPUB, AsciiDoc, LaTeX, CSV, audio and domain XML (USPTO, JATS, XBRL) with a unified pipeline API.
Understands multi-column scientific layouts, sidebars, headers, footers and visual hierarchy automatically.
State-of-the-art table structure extraction reconstructs complex merged cells and borderless financial tables.
Choose EasyOCR, Tesseract, RapidOCR, OcrMac or NVIDIA Nemotron OCR for scanned documents.
Chunk on structural nodes instead of character tokens, preserving heading hierarchy and tables.
Export lossless DoclingDocument JSON, Markdown or DocTags with integrations for LangChain, LlamaIndex, Haystack, CrewAI and MCP.
Docling integrates with leading frameworks and platforms.
| Tool / Extension | Category | Open Source | Best For |
|---|---|---|---|
| docling-serve | FastAPI REST Server | MIT | Self-hosting an HTTP document conversion endpoint for microservices. |
| docling-mcp | MCP Server | MIT | Connecting Docling to AI desktop apps like Claude and LM Studio. |
| langchain-docling | RAG Framework | MIT | Loading parsed documents directly into LangChain vector stores. |
| llama-index-readers-docling | Data Framework | MIT | Creating structured LlamaIndex document nodes from PDFs. |
| Apify Docling Actor (third-party) | Cloud Scraping | Pay per usage | Automating cloud-scale web document extraction. |
A deep, practical reference for the Docling architecture, pipelines and APIs.
Answers to common questions about Docling installation, features and troubleshooting.
Docling.org is independent. For authoritative information, always use the official project.