marker vs docling
Document conversion tools for structured AI workflows
marker and docling convert documents into structured content for search and downstream applications, and both can run locally. marker focuses on document conversion with configurable outputs and processing, while docling covers a wider range of input formats and offers a unified document representation.
marker: Convert Documents into Structured Text
Marker converts PDFs and other documents into Markdown, JSON, HTML, or chunks, preserving structure such as tables, equations, and images. It suits developers building document-processing workflows who can run its local models and inference backend.
docling: Parse Documents for AI Workflows
Docling converts documents from formats such as PDF, Office files, images, and audio into structured representations and exports. It suits developers building document-processing and generative AI pipelines that need format coverage, OCR, or local execution.
| marker | docling | |
|---|---|---|
| Language | Python | Python |
| License | Apache-2.0 | MIT |
| Stars | 40.2k | 68.3k |
| Forks | 2.9k | 5k |
| Last analyzed | Oct 3, 2026 | Oct 3, 2026 |
Key differences
- marker converts PDFs and, with additional dependencies, image, Office, HTML, and EPUB files; docling also lists audio, video, and email formats.
- marker outputs Markdown, JSON, HTML, or chunks; docling also exports WebVTT, DocTags, and DocLang, and uses a unified DoclingDocument format.
- marker uses Surya for OCR and layout processing through a local inference server, with balanced and fast modes; docling capabilities and parsing behavior vary by format and pipeline.
- marker is Apache-2.0 licensed, while docling is MIT licensed; both note that model licenses can be separate from the code license.
- marker includes a small-scale local API server; docling offers a CLI, Python API, MCP server, and API server.
- marker reports 40.2k stars and 2.9k forks; docling reports 68.3k stars and 5k forks.
Choose marker if you…
- need configurable conversion outputs, including chunked content, for a self-hosted workflow.
- want Python integration or a small local API server for document conversion.
- can run its local inference backend and have checked the separate model terms.
Choose docling if you…
- need to parse a wider range of formats, including audio, video, or email.
- want a unified document representation and integrations with AI frameworks such as LangChain or LlamaIndex.
- need options including an MCP server or local execution for sensitive or air-gapped workflows.
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.