Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

MinerU

opendatalab.github.io

  • Free plan
  • Open source
  • Windows
  • macOS
  • Linux
  • Web
  • API
  • Self-hosted

What MinerU is

MinerU 4.0 is a document parsing and conversion tool with a local document library and services for application integration and agent reading. It accepts PDFs, images, office documents, ebooks, web pages, and other listed formats, then exports Markdown, HTML, LaTeX, DOCX, EPUB, PDF, structured content, or content lists. Four parsing tiers—Flash, Basic, Standard, and Advanced—let users choose among different quality and compute needs. Its library can find files, cache parsing results, search content, resume work by page or block, and retain citation locators. Developers can connect through a Python SDK, V1 HTTP API, command-line tools, or a Gradio-based WebUI. MinerU also supports MCP and agent frameworks, with export options for Dify and Notion. Documents are not automatically uploaded to the official service; remote parsing requires explicit configuration. Local WebUI and API deployments are documented for Linux, macOS, and Windows. The open-source software is free under its project license; commercial use requires a separate license above 100 million monthly active users or USD 20 million monthly revenue.

Who it suits

MinerU suits people who need to parse and convert documents locally, or developers integrating document processing into applications and agent workflows. Its local deployments require suitable hardware; the docs recommend at least 16 GB system memory for higher-throughput use and at least 16 GB unified memory on Apple Silicon.

Good to know, and what to watch

  • Accepts a wide range of document and image formats.
  • Exports to Markdown, HTML, LaTeX, and other formats.
  • Library supports search and citation locators.
  • Provides SDK, HTTP API, CLI, and WebUI.
  • Remote parsing requires explicit configuration.
  • Higher-throughput local use calls for at least 16 GB system memory.
  • Commercial use above stated thresholds requires a separate license.
  • Third-party online services based on MinerU must provide attribution.

On-premise

—

Free

Free

MinerU open-source software

Free

Pro · Monthly

$9.90 one-time

Open source

Getting started

  1. Open https://opendatalab.github.io/MinerU/
  2. Choose a Linux, macOS, or Windows installation path
  3. Install the software for local, API, or self-hosted use
  4. Select a parsing tier for the quality and compute needs
  5. Provide a supported document input
  6. Export the result in a supported format

Limits to check

MinerU is free under its project license, but a commercial license is required above 100 million monthly active users or USD 20 million in monthly revenue. Third-party online services based on MinerU must provide attribution.

The verdict

MinerU brings document parsing, a searchable library, and multiple integration options into one open-source tool. Check the license thresholds and local hardware recommendations before choosing a deployment.

Questions people ask

Is MinerU free and open source?

Yes. The open-source software is free under its project license.

Which platforms can run MinerU?

It supports web, API, Linux, macOS, Windows, and self-hosted deployments.

Can MinerU process documents locally?

Yes. Local WebUI and V1 API server deployments are documented, and documents are not automatically uploaded to the official service.

What input formats does it accept?

Supported inputs include PDF, images, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, ODT/ODS/ODP, EPUB, OFD, HTML/MHTML, and CSV/TSV.

What outputs can it create?

Outputs include Markdown, HTML, LaTeX, DOCX, EPUB, PDF, structured content, and Content List V1/V2.

Does it provide an API?

Yes. MinerU provides a V1 HTTP API, Python SDK, command-line tools, and a Gradio-based WebUI.

What the maker publishes

Agent integrations
The official site says MinerU supports the MCP protocol and agent frameworks, and offers export to Dify and Notion. source
API network limitation
The hosted create-task API documentation warns that URLs hosted on GitHub, AWS and other overseas services may time out. source
API rate limits
The hosted service documents 300 submit requests per minute, 1,000 result queries per minute and up to 10,000 files per day, subject to dynamic adjustment. source
Changelog license change
Release 3.1.0 dated April 18, 2026 changed the project from AGPLv3 to the custom Apache-2.0-based MinerU Open Source License. source
Cloud lightweight API
The Agent Lightweight API requires no login but is IP-rate-limited, handles one file up to 10 MB and 20 pages, and returns Markdown only. source
Cloud precision API
The Precision Extract API requires a token, accepts files up to 200 MB and 600 pages, batches up to 200 files, and returns ZIP results with Markdown and JSON plus optional DOCX, HTML or LaTeX. source
Company headquarters and size
Crunchbase lists OpenDataLab as a private for-profit organization headquartered in Shanghai with 51–100 employees. source
Company identity
OpenDataLab's official site identifies it as the Data Platform Center of Shanghai Artificial Intelligence Laboratory and the maker of MinerU. source
Developer tooling
Official tooling includes a Python SDK, V1 HTTP API, stateless batch conversion, multi-service Router and Gradio WebUI. source
Docker support
The documented Docker workflow targets Linux NVIDIA GPUs and Windows WSL2; macOS Docker has no MPS acceleration and non-NVIDIA deployment is pending. source
Document library
The document library can discover files, cache parse results, search content, continue by page or block, and preserve citation locators. source
Document library
Its library discovers and caches files, supports content search and continuation by page or block, and preserves stable citation locators. source
Export formats
The document model supports Markdown, HTML, LaTeX, DOCX, EPUB, PDF, Structured Content and Content List V1/V2 outputs. source
Hardware requirements
The docs recommend at least 16 GB system memory for higher-throughput local deployments and at least 16 GB unified memory for Apple Silicon. source
Hosted service
The official site links to online use and describes an official cloud service, but the opened pages did not state its pricing. source
Input formats
The documentation lists PDF, images, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, ODT/ODS/ODP, EPUB, OFD, HTML/MHTML, and CSV/TSV as inputs. source
Input formats
Supported inputs include PDF, images, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, OpenDocument, EPUB, OFD, HTML/MHTML and CSV/TSV. source
Installation
The base package is installed with mineru>=4.0,<5 and includes the WebUI, ONNX runtime and llama.cpp VLM. source
Institution founding
LinkedIn lists Shanghai Artificial Intelligence Laboratory as founded in 2020, headquartered in Shanghai, with 201–500 employees. source
Integration
MinerU provides a Python SDK, V1 HTTP API, command-line tools, and a Gradio-based WebUI. source
License
MinerU uses a custom MinerU Open Source License based on Apache 2.0, with attribution requirements and a separate commercial-license trigger above 100 million monthly active users or 20 million USD monthly revenue. source
License terms
The project license permits commercial use without a separate license unless monthly active users exceed 100 million or monthly revenue exceeds USD 20 million, and requires attribution for third-party online services based on MinerU. source
Limit: active defect areas
The repository bug list includes reports involving Apple MPS OCR falling back to CPU, formula and punctuation errors, figure provenance, table extraction and Arabic recognition/order. source
Limit: backend quality gap
DocuOCR reports the CPU pipeline backend at 86.47 on the same benchmark, about 8.92 points below the VLM result. source

Checked against the maker's pages on 15 Oct 2026.