Your agents use your PC and browse any page.

Features

• All in one

  Browser and desktop in a single server.

• Cheaper than Computer Use

  89% fewer vision tokens: sends only what changed.

  Simulated 20-turn benchmark

• Standard MCP

  Any model. Tested on Claude Code.

• Browser in DOM mode

  Reads the page through its code, without images. For sites that allow automation.

  Full read: 24–34 ms

• Undetectable browsing mode

  Your everyday browser, with a real mouse and keyboard. For demanding sites.

  navigator.webdriver = false

• Native OCR

  Reads the text on screen without sending images to the agent.

• Capture zones

  Looks only at the part of the screen that matters, not the whole screen.

• Change detection

  Sends only what changed since the last look, and can wait for something to change.

• Action log and sounds

  Every action is recorded; optional sound cues.

• Action scripting

  PRO

  Record a desktop task as a Macro and repeat it with a single call.

• Mini button captures

  PRO

  Save a button once in Elements; the agent finds it and presses it.

• Also controls your remote PCs

  PRO

  Over an SSH tunnel, with your own keys.

Examples

• Paperwork on web portals
• Windows programs without an API
• Extracting data from web pages
• Testing your own application
• Changing Windows settings
• Repeating desktop tasksPRO

Install

1. Download the installer (Verdesk_x.y.z_x64-setup.exe) from GitHub Releases.
2. Run it. It doesn't ask for administrator rights; when it finishes, the Verdesk icon sits in the system tray.
3. Click the tray icon → Settings → Connection → Copy prompt, and paste it into your agent's terminal (Claude Code, Codex, OpenCode, etc.).
4. Close and reopen your agent's terminal so it loads the MCP server.