Preface¶

In DeepSeek Harness (DSH), agents typically need to convert document content into processable text. PDF, Word, PPT, Excel, HTML, and image files cannot be “read” directly by models alone; they are better parsed into structured Markdown first. dsh-mineru connects MinerU’s parsing capabilities as a DSH plugin, enabling agents to perform document parsing during tool calls. Below is an overview of its capabilities, installation, usage, and limitations.

What This Is¶

Lee-Hilex/dsh-mineru is an MIT-licensed DSH plugin maintained by Lee-Hilex. It uses MinerU as the parsing backend to convert PDF, Word, PPT, Excel, HTML, and image files into structured Markdown. The plugin provides two API paths: the Precision API (used when a token is configured) and the tokenless Agent API (used when no token is configured).

Core Features¶

• Multimodal document parsing: Supports PDF, Word, PPT, Excel, HTML, and images, outputting structured Markdown.
• Dual API modes: Precision API requires a MinerU token; Agent API requires no token but is IP rate-limited.
• Automatic selection: API mode is automatically chosen based on token configuration, with per-call override support.
• Progressive tool exposure: mineru_activate is exposed first; after activation, mineru_parse, mineru_batch_parse, and mineru_task become available.
• Web UI upload: After drag-and-drop file upload, file paths are filled into the composer draft.
• Batch parsing: Precision API supports automatic chunked batch parsing.
• Rate limiting handling: Includes token buckets, daily caps, 429 backoff, and actionable error messages.
• Results persisted to disk: Parsing results are stored as workspace files, with signed preview links provided in the Web UI.
• Other entry points: Supports URL document parsing and headless CLI usage.

API Modes and Limitations¶

Installation and Activation¶

First, confirm the environment: DeepSeek Harness 0.1.0-rc.6 or a compatible 0.1.x version is installed, and the dsh CLI is available.

Below is the installation command for a Web profile:

dsh plugin --profile web add dsh-mineru

After installation, you can confirm the plugin is registered with the following command:

dsh --profile web --dump-config | grep -A 2 "id: mineru"

If a Web UI is configured, you can open:

Settings → Plugins → MinerU Parser

Token configuration is optional.

To configure a token in the Web UI:

Settings → Plugins → MinerU Parser

Enter the token in the write-only field; leaving it blank will preserve the existing value.

For headless or CLI usage, you can use:

$DSH_HOME/.credentials.yaml

or export:

MINERU_API_TOKEN

The token is resolved on each operation; rotated tokens take effect on the next call without requiring a restart.

If installing from a local source, use the packaged tarball. Do not use pnpm link:, as it resolves @deepseek-ai peers outside the profile.

Typical Usage¶

1. Web UI drag-and-drop parsing: Drag a PDF, Word document, or image into the chat, then enter, for example:

Parse this PDF to Markdown and summarize the key points

After sending, preview the returned file.

2. URL parsing: Have the agent process:

parse this document: https://example.com/paper.pdf

3. Headless smoke test:

dsh --profile headless "Parse C:/docs/sample.pdf to Markdown and summarize the first 3 paragraphs"

4. Verify installation:

dsh --profile web --dump-config | grep -A 2 "id: mineru"

Suitable Scenarios and Considerations¶

This plugin is suitable for DSH workflows that require reliable document reading, batch parsing, URL parsing, or headless invocation. Usage considerations:

• Without a token, only the Agent API is available, which has smaller limits and is IP rate-limited; for HTML, .doc / .ppt / .xls support, or batch parsing, configure a token to use the Precision API.
• The plugin runs with the current dsh process permissions, reads workspace files, calls the MinerU API, and generates workspace files; review the source code and MIT license before installation.
• package.json declares web client injection slots as slots, connection, sessions, and conversation, and depends on @deepseek-ai/cordis ^4.0.1, @deepseek-ai/schemastery ^3.18.0, and @deepseek-ai/dsh-* ^0.1.0-rc.6 related packages.
• The community directory is an independent site with no official affiliation to DeepSeek / High-Flyer; refer to the repository and verified materials for specific plugin information.

Conclusion¶

The value of dsh-mineru lies in connecting MinerU’s document parsing capabilities to the DSH plugin ecosystem: models do not need to directly handle binary files, yet can still obtain structured Markdown from PDF, Office, HTML, and image files through tools. GitHub: https://github.com/Lee-Hilex/dsh-mineru. Directory page: Verified materials do not provide a confirmable directory page URL, so search for Lee-Hilex/dsh-mineru in the plugin directory.