Assign a Desktop to an Agent
Assigning a cloud desktop to an agent gives it the ability to interact with graphical applications using screenshots and mouse/keyboard actions -- powered by Claude computer use.
Prerequisites
- A running agent
- A provisioned cloud desktop (see Provision a Cloud Desktop)
Steps
1. Open the Desktop tab
Navigate to Agents in the sidebar, click your agent, then select the Desktop tab.
2. Select a desktop
Choose an available desktop from the dropdown. The list shows all provisioned desktops in your project along with their current status.
3. Save the assignment
Click Save. The agent now has access to the desktop.
How it works
Once a desktop is assigned, the agent gains access to a desktop automation tool. When the agent uses this tool, it follows a screenshot-action loop:
- The agent takes a screenshot of the desktop.
- It analyzes the screenshot to understand what is on screen.
- It performs an action (click, type, scroll, or keyboard shortcut).
- It takes another screenshot to verify the result.
- It repeats until the task is complete.
This loop uses Claude's computer use capability to operate any application with a graphical interface.
Verify
- The Desktop tab shows the assigned desktop name and status.
- Ask the agent to perform a task on the desktop (e.g., "Open the browser and go to example.com"). The agent should describe what it sees and the actions it takes.
The agent can only access one desktop at a time. To switch desktops, return to the Desktop tab and select a different one.
Desktop tasks work best with clear, specific instructions. Instead of "check my email," try "open Firefox, go to gmail.com, and summarize the three most recent unread emails."