Who it is forA knowledge worker who wants an agent to gather search results in a browser and report back, with a check that the results actually loaded.
The problemBrowsing and copying results by hand is slow, and an agent that does not check the results page may report a search that never happened.
The outcomeThe agent opens the browser, enters the query, confirms the result page, and returns a summary of its actions and what it verified.
What you need first
- A Windows computer with DSH Desktop installed
- Windows PowerShell 5.1 (included with Windows)
- For DSH Web or a standalone MCP client: Node.js 22 or later
- A logged-in desktop session that the plugin can use
Steps
- Run
dsh plugin --profile desktop add dsh-computer-use-winin the DSH terminal opened from the tray icon. - Quit DSH Desktop from the tray and reopen it so the new plugin loads.
- Open Settings → Computer Use to check the tool loading option, then leave it on the default.
- Give the agent the search query and the text that must appear on the results page.
- Read the final summary and confirm the agent named what it saw on the results page.
Prompt for your agent
Using the dsh-computer-use-win tools, open a browser, search for <query>, and confirm the results page shows <expected text>. Summarize what you did and what you verified in no more than five sentences. Do not report success unless the expected text was observed.
Caveat. Chromium and Electron apps can ignore background input, so bring the browser window to the front before the agent types, and results depend on the page and the model used.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for Yu-tao-Li/dsh-computer-use-win and "Run a browser search and have an agent summarize results"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer