MinerU

What is MinerU?

MinerU is an intelligent document parsing tool designed for document digitization, knowledge bases, and AI applications. It extracts text, tables, and formulas from PDFs, scanned documents, images, and common Office documents, organizing them into structured content that is easy to read and process further. For users who frequently encounter issues such as “incorrect text order in two-column papers,” “unsearchable scanned documents,” or “distorted tables after pasting,” MinerU offers a more comprehensive solution than simply copying and pasting.

The official website offers online services, client software, and API access, and is integrated with open-source projects. Individual users can start by processing sample files via the web interface; teams that need to organize data in bulk, build a RAG knowledge base, or integrate with Agent workflows can further evaluate the API and on-premises deployment options. Here, RAG refers to a system where AI first retrieves information from its own data and then uses that information to answer questions.

Key Features

1. Recognizing Complex Document Structures:Analyze content such as headings, paragraphs, images, and multi-column layouts to help restore a logical reading order and reduce the need for subsequent manual cleanup. Scanned documents can be processed using OCR; however, the results should still be cross-checked against a sample of the original pages.
2. Working with Tables and Mathematical Formulas:Organize tables into a structure that can be further utilized, and convert formulas into expressions such as LaTeX. For research papers, technical reports, and documents containing a large number of parameters, this is more valuable than simply extracting a long string of text.
3. Output machine-readable content:Common output formats include Markdown and JSON. Markdown is suitable for reading, editing, and importing into knowledge bases; JSON is better suited for further processing by programs based on information such as paragraphs, content blocks, or positions. Export capabilities vary by entry point and version; please refer to the actual interface documentation for details.
4. Compatible with a variety of document sources:The official website lists options such as PDF, Word, PPT, Excel, images, and web page URLs. Before uploading, you should verify that the selected mode supports the corresponding format; do not assume that every online option offers the full range of capabilities available in the open-source component.
5. Integrating Automated Workflows:Provides a document parsing API and demonstrates integration methods for MCP and Agent. Teams can streamline the process of “collecting, parsing, validating, and storing files” into a single workflow, eliminating the need for repetitive manual operations for each file.

core value

The most valuable feature of MinerU is its ability to transform documents—which are difficult to use directly—into data that can be processed by subsequent tools. Researchers can organize literature, operations and consulting staff can compile reports, and developers can prepare data for knowledge bases. It is particularly well-suited for situations where “documents already exist, but the information within them is difficult to extract.”

The quality of the analysis depends on the clarity of the source document, the language, the layout, and the model selected. Small text, low-quality scans, cross-page tables, and complex formulas require special attention during review. Even after document extraction is complete, this does not guarantee that citations are correct, numbers are reliable, or that AI-generated answers will necessarily be accurate. It is recommended to retain the correspondence between the source document and page numbers to facilitate future verification.

Usage costs should also be considered separately by entry point. As of October 11, 2026, the official website’s API documentation specifies that individual files must not exceed 200 MB or 200 pages, with a daily quota of 1,000 pages per account at the highest priority level; priority is reduced once these limits are exceeded; The Agent Lightweight Parsing API allows a maximum of 10 MB per file and 20 pages, with IP rate limits. These restrictions apply to the specified APIs and do not represent a uniform quota for the web version or unlimited free usage. Specific pricing and available quotas for current online services should be confirmed at the actual point of use.

For documents subject to confidentiality requirements, review the authorization and terms of service before uploading them to the cloud. For on-premises deployments, verify the model, operating mode, and connections to external services. Open-source projects currently use the MinerU open-source license with certain conditions; read the license before commercial integration, and do not assume there are no usage restrictions simply because the software is “open source.”

How to use

1. First, clarify your goals:I just want to organize my reading notes, so I’ll prioritize evaluating Markdown; if I need to import data in bulk or preserve structural information, I’ll take a closer look at JSON and related resources.
2. Select an entry point from the official website:Upload an authorized sample file online or via the client. The initial test should include body text, tables, and formulas that you are likely to encounter in practice.
3. Check key pages:First, check the reading order, table headers, merged cells, numerical units, formula symbols, and image mappings; then decide whether to expand the batch.
4. Exported and organized by:Keep the original file, the analysis results, and any necessary image resources; note the source page numbers for important content; and then import them into your notes or knowledge base.
5. Batch Pre-Validation Interface:Select the Precision API, Lightweight API, or deployment solution based on file size, number of pages, frequency, and privacy requirements, and handle queuing, retries for failed requests, and result validation.

Comparison of similar products

If your goal is to annotate PDFs, directly modify page layouts, or create finished documents with precise typesetting, you’ll also need to use the appropriate PDF readers, editors, or typesetting software. When selecting a document parsing tool, comparing results using your own actual samples is more helpful than simply looking at the accuracy rates listed in promotional materials.