Skip to content

Windows Computer Use for DSH

by Yu-tao-LiConnectorGet organised

Lets an AI agent see and use Windows apps and web pages: click, type and read the screen

View on GitHub

About Windows Computer Use for DSH

Windows Computer Use for DSH is a connector that lets an AI agent see and use Windows programs and web pages: find buttons, read on-screen text, click, type and manage windows.

For Windows usersDeepSeek Harness users

Made by Yu-tao-Li

An open-source project by GitHub user Yu-tao-Li, made as a plugin for DeepSeek Harness Desktop and usable in other MCP clients.

Stars
★ 355
License
MIT
Cost
Free
Level
Intermediate
Runs
Local

Screenshots

From the project's own repository · 3 images, swipe or use the arrows.

An agent filling in a sample expense form in the DSH browser, with a pointer marking its click
Conversation rows listing each desktop action, such as key presses and result checks
The Computer Use settings page with browser options and tool loading presets
Image: Yu-tao-Li (MIT). Original file

Overview

This connector lets an AI agent work with the real Windows programs and web pages on your computer. It finds buttons and text fields through Windows UI Automation, reads on-screen text with the built-in Windows OCR, clicks, types and manages windows, and works on web pages in a background browser without taking your mouse. It is made as a plugin for DeepSeek Harness (DSH) Desktop and also runs in other MCP clients. It suits Windows users who want an agent to carry out desktop chores.

It works in the background where it can, shows clear on-screen signs when it takes the mouse, and stops when you press Esc.

Try these first

Start with the first one: it is a small task you can check easily.

  1. Type “Meeting notes” into Notepad, then read the text back to confirm it was entered.First try · types on desktop

  2. Calculate 25 × 4 in the open Calculator window and read the result.

  3. Open a browser, search for “Windows UI Automation”, confirm that results appeared and summarize what you did.

  4. Fill in the form at [page] with this sample data and stop before submitting: [data]fills a form

Example prompts written by Agent Flows Hub.

Capabilities

From the creator's documentation.

ToolsetWhat your agent can doAccess
Windows and controlsLists open windows and finds buttons and fields with UI AutomationRead
Screen textReads on-screen text with the built-in Windows OCRRead
Click and typePresses buttons, fills fields, types text and sends key pressesRead + write
Browser laneReads and fills web pages in Chrome or Edge in the backgroundRead + write
Window managementSwitches, moves and manages windows on the desktopRead + write
Wait and checkWaits for a window, control or text, then confirms the resultRead

Good to know

You want an agent to do a short task in a Windows app, such as typing into Notepad or searching in a music app. The connector finds the controls, acts on them and reads the screen to check the result.

Limits

  • Screenshots and recognized text can include private content from your windows; whether they reach a remote model depends on your DSH and model settings.
  • Windows only. The settings page, step records and on-screen indicators exist only inside DSH; other MCP clients get the tools alone.

Notes by (original project name: dsh-computer-use-win), written from the creator's documentation, not from running it ourselves. dsh-computer-use-win README (English)

Stuck setting it up?

Tell us what happened. We take a look first; if it needs paid work, you get an estimate before anything starts.

Get help

Other connectors

  • Wix MCP Lets your AI assistant manage your Wix sites, from store products to bookings and blogs
  • Xiaohongshu MCP Lets your AI agent search Xiaohongshu and post picture or video notes from your account
  • Zapier MCP Lets your AI agent send emails, add rows and post messages in 9,000+ apps
  • Zotero MCP Lets your AI assistant search, read, summarise and annotate papers in your Zotero library
  • Airtable MCP Lets your AI agent find, add and update records in your Airtable bases
  • AntV Chart MCP Lets your AI agent turn your numbers into charts, mind maps and flow diagrams

See all 52 connectors · An MCP server is missing or disconnected

GitHub