Give the agent a desktop of its own.
You want your agent to use a graphical program without touching your screen or seeing your other windows.
Run graphical tests where there is no screen.
Your agent runs on a server or in a container, with no display and no graphics card, and it has to check something that only works in a real graphical program.
Open an app on it and watch.
You want a graphical program running somewhere it cannot disturb you, with a picture of what it shows.
The captured image shows the desktop with the browser window open on the documentation page.
Check a page in two browsers and at two sizes.
You changed a web page and want to see how it looks in more than one browser, and what happens when the window gets smaller, without opening anything on your own screen.
Change the desktop size while it runs.
You want to see how something looks at another size, or the first size was a guess.
Freeze the desktop, then carry on.
You want the agent to stop working for a while without losing what is open, then continue exactly where it was.
Find a session again after a crash.
The server stopped without a clean shutdown, or a colleague released a desktop, and you want to finish it properly instead of hunting for processes.
Record the desktop with sound and picture.
You want proof of what an application did on its desktop, with the sound it played, so a person can watch and listen to it afterwards. Or you are testing an application that listens to a microphone.
To test a microphone instead, ask the agent to play a prepared audio file into the desktop and report how much was sent. See the MCP RDP tools for the recording and sound options.