Lee-Hilex/dsh-mineru
A MinerU-powered multimodal document parsing plugin: turn PDF, Word, PPT, Excel, HTML and images into structured Markdown (tables, formulas and images preserved) so text-only models can read documents.
dsh-mineru plugs MinerU, OpenDataLab's high-precision document parsing engine, into the DSH agent ecosystem so text-only models can read documents: the model receives structured Markdown with tables, formulas and images preserved instead of a binary file it cannot open. It parses PDF, Word (.doc/.docx), PPT, Excel, HTML and common image formats, with the Agent API limited to the modern Office formats and no HTML (the Precision API handles the legacy formats and HTML). Highlights include dual API modes with per-call override, progressive tool exposure (only a tiny mineru_activate bootstrap is visible until the agent activates the full toolset), drag-and-drop uploads that place a file path in the composer draft, auto-chunked batch parsing that respects official limits, and rate-limit-aware token buckets with 429 backoff and actionable error messages.
Install
dsh plugin --profile web add dsh-mineruCopied verbatim from the README's Quick start (Web profile). The README also shows a headless form and a local-checkout form.
Compatibility
A DSH plugin that speaks to MinerU's hosted parsing service (OpenDataLab). Two API modes are auto-selected: the Precision API needs a MinerU token, while the tokenless Agent API is IP rate-limited. The README states plainly that there is no fully offline mode — every parse uploads the file to MinerU's remote service, so it must not be used for confidential material.
Details
- Repo: Lee-Hilex/dsh-mineru
- Category: Tools & Capabilities
- Stars: 6
- Version: GitHub repo Lee-Hilex/dsh-mineru; npm package dsh-mineru at 0.1.13 (registry check 2026-10-09; registry repository field resolves back to this repo). MIT license; targets DeepSeek Harness 0.1.0-rc.6+.
Alternatives
Aidenwu0209/dsh-PaddleOCR-Skills · aaronlei/dsh-web-search-diy · anweat/dsh-web-search-pro