Give your AI Agent sharp eyes — An Agent-Native Document Reading Skill (Preview Version)
Faced with piles of PDFs, Agents often "brute-force" documents page-by-page to extract data—a process that is inefficient and costly in tokens. MinerU Document Explorer solves this by providing AI Agents with human-like reading skills. Through four modular "atomic" capabilities, it enables flexible extraction and navigation of PDF.
| Capability | Description |
|---|---|
| Logic Retrieval — understand index, know structure | Navigates the document outline and jumps directly to the target section — no more scrolling from page one |
| Semantic Retrieval — understand meaning, ignore exact wording | Natural language queries that find the right page even when you don't know the exact terms; works cross-language |
| Keyword Retrieval — precise positioning, one-click match | Regex-powered full-document scan that flags every occurrence of a name, ID, or pattern |
| Evidence Extraction — text & image combined, cut and paste | Precisely crops tables, figures, formulas, and other fine-grained elements with element-level citation metadata |
After integrating MinerU Document Explorer, on our benchmark tasks:
- 📉 ~40% fewer tokens (tested on Claude Opus 4.6): average dropped from 45k to 28k tokens per task
- 🎯 20%+ higher task success rate (tested on Minimax 2.1): from 60–70% up to ~90%
Note: The demo video is currently in Chinese. English version coming soon.
demo_poster.mp4
Features demonstrated: Logical Retrieval | Semantic Search | Keyword Search | Evidence Extraction
Real-world use cases:
- 🏦 Act as a "Financial Analyst" — Auto-generate research report PPT
- 📚 Act as a "Teaching Assistant" — Grade CET-6 exams and create error notebooks
With only one line of code, you can install our skill.
clawhub install mineru-document-explorer
After successful execution, Send the following to your OpenClaw agent:
Install and test mineru-document-explorer
Send the following to your OpenClaw agent — it will handle everything automatically:
By using `git clone`, install this PDF reading skill for me: https://github.com/opendatalab/MinerU-Document-Explorer, then test it.
The agent will: read the install guide → copy the skill directory → run the setup script → guide you through configuration.
After setup, the agent will ask if you want to configure PageIndex (optional — provide an OpenAI-compatible API key to enable auto-generated document outlines; skipping this does not affect any other feature).
If this project helps your work, please cite:
@article{wang2026agenticocr,
title={AgenticOCR: Parsing Only What You Need for Efficient Retrieval-Augmented Generation},
author={Wang, Zhengren and Ma, Dongsheng and Zhong, Huaping and Li, Jiayu and Zhang, Wentao and Wang, Bin and He, Conghui},
journal={arXiv preprint arXiv:2602.24134},
year={2026}
}
@article{niu2025mineru2,
title={Mineru2.5: A decoupled vision-language model for efficient high-resolution document parsing},
author={Niu, Junbo and Liu, Zheng and Gu, Zhuangcheng and Wang, Bin and Ouyang, Linke and Zhao, Zhiyuan and Chu, Tao and He, Tianyao and Wu, Fan and Zhang, Qintong and others},
journal={arXiv preprint arXiv:2509.22186},
year={2025}
}Thanks to PageIndex for the hierarchical document outline construction capability that powers the Logic Retrieval feature in this project.
Thanks to Qwen3-VL-Embedding for providing multimodal document reranking capabilities, which provided important support for our semantic search.
This project is open-sourced under the MIT License.