MIT-licensed Windows-only DSH native plugin that gives the text-only DeepSeek eyes and ears: image recognition / OCR / object detection / video understanding / speech transcription / document visualization, one-click install.bat, depends on a cloud vision model (default Qwen).
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-03
License
MIT
01What can it help you accomplish?
Give DeepSeek Harness multimodal sight and hearing — describe, OCR, detect, understand video, transcribe speech, read documents
Image Q&A, extracted text, bounding boxes, video summaries with audio timeline, transcripts, and PDF/Office/CSV/code document answers — straight in the DeepSeek chat
DeepSeek Harness (dsh) users on Windows who want their text-only model to see images, hear audio and read documents
Read a screenshot the moment you capture it, without saving and dragging a file
The pasted clipboard is auto-replaced with the screenshot PNG path so DeepSeek reads the image directly
Windows users who ask DeepSeek questions about screenshots and want zero manual file handling
02How to install into DeepSeek Harness
Prerequisites
- Windows + PowerShell (the install needs Windows; ~5 minutes total)
- A cloud multimodal model API key (default Qwen / DASHSCOPE_API_KEY) — required, the plugin has no model of its own
- `uv` must be installed (the install self-check warns if missing; MCP servers depend on it)
Installation steps
- 01
Clone or download the repo: `git clone https://github.com/Yts1919/dsh-vision-complete.git` (or the Gitee mirror), then unzip to any folder (e.g. C:\DeepSeek-Vision).
$ git clone https://github.com/Yts1919/dsh-vision-complete.git
- 02
Double-click `install.bat` in the folder — it installs the skill, registers the MCP servers, installs the screenshot tool and runs a self-check. (Or run `powershell -ExecutionPolicy Bypass -File .\install.ps1`.)
- 03
Set your cloud vision API key: `setx DASHSCOPE_API_KEY "sk-你的key"` (get one from Alibaba Cloud Bailian), then fully quit and restart DeepSeek Harness.
- 04
Verify: drag any image into the DeepSeek chat and say "describe this image".
Verify the integration
- Find any image (or press Win+Shift+S to screenshot) and drag it into the DeepSeek chat box.
- Say "describe this image" to it.
- ✅ A returned image description = installed; ❌ a "no API Key / 401" error means go back to step 3 and confirm DASHSCOPE_API_KEY is set and DeepSeek Harness was fully restarted.
Rollback
- Double-click `uninstall.bat` in the folder to remove the plugin.
03DSH integration and capability boundaries
Native DSH plugin: installs the vision-multimodal skill into ~/.dsh/skills and registers the qwen-mm-plugins MCP servers (core / api / video-memory) into DSH; Windows-only install via install.bat.
Image understanding & Q&A
image (dropped into chat or a screenshot path)→description, Q&A, image captioning
OCR text extraction
screenshots, receipts, document photos→extracted text
Object detection & localization
image→target location, bounding box, crop
Video understanding
video→frame+audio timeline, event search, counting
Speech transcription
audio / meeting recording→transcript, with timestamps, speaker diarization
Document visualization
PDF / Office / CSV / code / 3D / notebook→document Q&A and visualization
Screenshot direct-read
Win+Shift+S screenshot→clipboard replaced with image path, DeepSeek reads it directly
While the screenshot monitor runs, the clipboard is replaced with the screenshot PNG path text
04Who is it for? When not to use it?
Good for
- DeepSeek Harness (dsh) users on Windows who want their text-only model to see images, hear audio and read documents
- Windows users who ask DeepSeek questions about screenshots and want zero manual file handling
Not for
- The plugin has no vision model of its own — you must supply a cloud multimodal model API key (default Qwen / DASHSCOPE_API_KEY); without a key it cannot be used even after install.
- Windows + PowerShell only; non-Windows cannot use it. Install takes about 5 minutes.
- The plugin calls a cloud vision model (default Qwen) to see/hear for you, so it needs network access and the vendor API key.
- API keys and other env vars only take effect after fully quitting and restarting DeepSeek Harness; MCP servers also load only after a restart.
05Compatibility, maintenance and safety notes
- The plugin has no vision model of its own — you must supply a cloud multimodal model API key (default Qwen / DASHSCOPE_API_KEY); without a key it cannot be used even after install.
- Windows + PowerShell only; non-Windows cannot use it. Install takes about 5 minutes.
- The plugin calls a cloud vision model (default Qwen) to see/hear for you, so it needs network access and the vendor API key.
- API keys and other env vars only take effect after fully quitting and restarting DeepSeek Harness; MCP servers also load only after a restart.
MIT · no tagged release (latest_release is null); last push 2026-08-16
06Frequently asked questions
How do I install dsh-vision-complete for DeepSeek Harness?
On Windows, unzip the repo and double-click `install.bat` (or run `powershell -ExecutionPolicy Bypass -File .\install.ps1`). It installs the vision-multimodal skill into ~/.dsh/skills, registers the qwen-mm-plugins MCP servers, and installs the screenshot tool.
Why do I need an API key?
The plugin has no vision model of its own — it calls a cloud multimodal model (default Qwen via DASHSCOPE_API_KEY). Without the key it can neither see nor hear; after `setx DASHSCOPE_API_KEY "sk-..."` you must fully restart DeepSeek Harness.
How do I verify the install in DeepSeek Harness?
Drag any image (or Win+Shift+S screenshot) into the DeepSeek chat and say "describe this image". A returned description means it works; a 401 means the key is unset or Harness was not restarted.
The model says it has no `mcp__qwen-mm-plugins-*` tool — what now?
Confirm `uv` is installed (the install self-check warns otherwise) and fully quit and reopen DeepSeek Harness; MCP servers only load after a restart.
How do I uninstall it?
Double-click `uninstall.bat` in the folder to remove the plugin.
07Related DSH workflows
modlens
by liustack
Vision bridge for text-only models: paste an image, get structured JSON evidence (OCR, layout, semantics).
agent-vision-toolkit
by anionex
Vision toolkit and skills that give text-only LLMs eyes — multi-image understanding, image Q&A, OCR, frontend UI restoration and GUI automation, with optional agent integration.
dsh-vision-router
by ysr666
Eyes for text-only DeepSeek Harness agents: built-in free vision chain (no key) + pixel-level vision tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots). One-command install, no Python, image turns work like ordinary tool-calling turns.
dsh-vision-toolkit
by anionex
为纯文本 DSH Agent 提供 10 个结构化视觉工具：意图感知图片问答、长截图 OCR、原始像素 grounding、UI 还原、像素 diff 等
08Data and sources
C:\Users\你的用户名\.dsh\skills\vision-multimodal\references\api-key-setup.md
**qwen-mm-plugins MCP**（安装时自动注册）
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-03. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.