Overview
This section explains how to connect to and operate the WorkSpaces Applications Model Context Protocol (MCP) server, the endpoint your agent uses to work in a desktop session. For a conceptual overview of WorkSpaces for AI agents and agent access, see Why WorkSpaces for AI agents? and How it works.
An agent works with the MCP server in four stages:
-
Authenticate. The agent authenticates to the MCP server before it can connect.
-
Initiate the connection. The agent opens a session and retrieves the list of available tools.
-
Make tool calls. The agent calls tools to operate the desktop, such as moving the mouse, typing, or taking a screenshot.
-
End the session. The agent ends the session when the task is complete.
Throughout these stages, WorkSpaces Applications handles the connection between the MCP calls and the underlying desktop. The topics in this section describe each stage. To authenticate and open a connection, see Connecting to the MCP server. For the tools your agent can call, see Available tools. To end a session, see Session cleanup.
An agent can connect to a streaming session for a stack, or to an image builder. To connect to either, see Connecting to the MCP server.
You enable agent access when you create a stack (using the WorkSpaces Applications console, AWS CLI, or API) or an image builder (using the AWS CLI or API). Before you connect an agent, review Prerequisites.
