Windows desktop GUI tools for the model: list windows with physical-pixel rects, activate one, screenshot a window, region, monitor or the whole desktop, then click, paste Unicode text and send key chords. Dependency-free PowerShell helper, per-monitor v2 DPI aware.
Install
# from npm (prebuilt)
dsh plugin --profile web add dsh-win-gui
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:alotherss/DeepSeekHarness-plugin#path:/packages/dsh-win-gui
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
Windows desktop GUI tools for the model: screenshot -> look -> click -> look again.
Six tools: win_gui_windows (window list with physical-pixel rects), win_gui_focus
(activate a window), win_gui_shot (capture a window, region, monitor or the whole desktop),
win_gui_click (click a physical pixel), win_gui_type (paste text, Unicode included), and
win_gui_key (send a chord such as ctrl+s).
It is for software that has neither a CLI nor an accessibility interface: old MFC tools, vendor flashing utilities, GUI-only firmware burners. If a CLI exists, use the CLI - this is not the first choice.
Tools
| Tool | What it does |
|---|---|
win_gui_windows |
Every titled window with its process, pid and physical-pixel rect, plus the monitor layout |
win_gui_focus |
SW_RESTORE + foreground, returns the rect. Call before win_gui_type or win_gui_key |
win_gui_shot |
Capture a window (default), a region, a monitor, or all; optional max_width downscale |
win_gui_click |
Absolute physical-pixel click, with right button, double-click, and a post-click wait |
win_gui_type |
Clipboard + Ctrl+V, so any Unicode text works; the previous clipboard is restored |
win_gui_key |
Key chords: enter, esc, tab, f5, ctrl+s, alt+f4 |
The coordinate model (where this normally goes wrong)
- The helper process declares per-monitor v2 DPI awareness before touching any window, so every coordinate it reports or accepts is a physical pixel, never DPI-virtualised.
- Window rects come from DWM extended frame bounds, not
GetWindowRect. On Windows 10 and later the latter includes the invisible resize border (about 7-8 px), which makes a capture slightly larger than the visible window and shifts every click. - Click coordinate = the capture origin (
left/top) + the offset inside the image. - Captures are downscaled to
max_width: 1600by default: a 4K window screenshot is roughly 5k tokens, while 1600 px stays readable at about 1.6k. The result reportsscaleand the original size, so a point measured on the image maps back asleft + round(imageX * scale),top + round(imageY * scale).
The loop
win_gui_windows --filter burner find the exact title and rect
win_gui_focus --window "Burner v2.1" activate
win_gui_shot --window "Burner v2.1" look; the image enters the context
win_gui_click --x 1373 --y 1132 click what you measured
win_gui_shot --window "Burner v2.1" verify, then adjust and repeat
Install
Inside DSH: Plugins -> Plugin market, search for dsh-win-gui.
Command line:
dsh plugin --profile web add dsh-win-gui
Requirements
- Windows. Any other platform fails with a clear message instead of pretending.
- PowerShell 7 (
pwsh) or Windows PowerShell 5.1 (powershell), tried in that order. SetDSH_WIN_GUI_POWERSHELLto a full path to pin one. - An interactive desktop session: captures use
CopyFromScreen, so a service or a session with no desktop returns nothing. - No dependencies at all: plain PowerShell plus .NET and user32. Nothing is installed or downloaded, and no browser engine is involved.
Notes
win_gui_typeandwin_gui_keydo not activate a window. Callwin_gui_focusfirst, otherwise the input goes to whatever happens to be in front.- A window capture activates that window first so nothing covers it. Pass
activate: false(the script's--no-activate) when focus must not move. - Every screenshot enters the context: prefer one window, then a
region;allis the most expensive option. - Multi-monitor coordinates are virtual-desktop absolute; a monitor to the left means negative X.
- Moving, resizing or changing the resolution of a window invalidates every coordinate you measured - enumerate or capture again.
Privacy and risk
A screenshot can contain anything on that screen (and all captures every screen). Those images
go into this conversation's context and the local temp directory only; the plugin itself makes
no network calls and uploads nothing. Clicks and keystrokes really act on your desktop:
capture first, confirm the target, and only then act.
Known limits
- Screen coordinates only. There is no accessibility tree, no template matching and no OCR - the model reads the picture.
- No UI-stability wait: a screenshot taken immediately after a click may catch a transitional
state; pass
wait_mswhen that matters. - Input goes through
SendKeysandmouse_event, which a few games and anti-cheat drivers ignore.
License
MIT
Links
More in this category
Tencent/WeKnora#dsh-weknora★ 37546
Four read-only tools over a WeKnora knowledge base: list knowledge bases, hybrid passage search, reassemble one document's chunks in order, and WeKnora's own cited RAG or ReAct-agent answer with a resumable session id.
DayuanJiang/next-ai-draw-io#dsh-plugin★ 36997
Draws draw.io diagrams in DeepSeek Harness through the next-ai-drawio MCP server: the link in the reply opens a live preview in the right sidebar, you can edit the diagram by hand, and each AI change is a version you can restore. Node.js 18+; npx downloads the MCP server on first use.
superdesigndev/treg★ 8164
Tool catalog for agents: search ~2,600 external endpoints (SEO and SERP, backlinks, social, people and company enrichment, ad libraries, scraping) by the task you want done, read each one's parameters and per-call price, then call it with the credential injected server-side. Ships the skill plus an MCP row that stays disabled until TREG_TOKEN is set.
T-Auto/dsh-ops★ 5864
Provides bash, PowerShell 7, and high-performance Rust tools for DSH on Windows to reduce token usage.
sixtysevenlf/dsh-blender-plugin★ 2147
Lets a DSH agent drive a local Blender over a direct TCP channel with 15 tools: viewport frames, custom-angle renders, Python execution, headless and job offload, transaction rollback, and a planning layer of 228 ops across 34 families.
helloHub1212/dsh-windows-mcp★ 1456
Bridges the CursorTouch/Windows-MCP desktop automation server into DSH through the official MCP client bridge, registering its screenshot, UI tree snapshot, mouse and keyboard input, clipboard, process and registry tools as mcp__windows__* tools for the model, and adds a windows_mcp_doctor diagnostic tool. Computer Use is off by default and is switched on from the plugin panel.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.