AWS appstream2 documentation change
Summary
Added definitions for agent session, computer input, vision and MCP service
Security assessment
Documents security aspects of new MCP service including IAM authentication, CloudTrail logging, and screenshot storage. Defines security-relevant capabilities but doesn't address a specific vulnerability.
Diff
diff --git a/appstream2/latest/developerguide/what-is-concepts.md b/appstream2/latest/developerguide/what-is-concepts.md index 405c12124..1e7a8ff4b 100644 --- a//appstream2/latest/developerguide/what-is-concepts.md +++ b//appstream2/latest/developerguide/what-is-concepts.md @@ -71,0 +72,20 @@ Make sure you are using latest WorkSpaces Applications images for multi-session +**agent session** + + +An _agent session_ is a WorkSpaces streaming session initiated by an AI agent. You create agent sessions by using the `CreateStreamingURL` API. The agent passes the streaming URL to the MCP service as a header on each request. You can monitor agent sessions through AWS CloudTrail, and you can view operational metrics in CloudWatch. You can optionally store screenshots captured during agent sessions in a customer-managed Amazon S3 bucket. + +**computer input** + + +_Computer input_ is a capability that allows agents to click buttons, enter text, and scroll on the desktop during a WorkSpaces streaming session. Computer input requires that you also enable computer vision. + +**computer vision** + + +_Computer vision_ is a capability that allows agents to see the desktop by taking screenshots during a WorkSpaces streaming session. + +**Model Context Protocol (MCP) service** + + +The managed _MCP service_ (`agentaccess-mcp`) provides AI agents with tools to click buttons, enter text, scroll, and take screenshots of the desktop during a streaming session. Agents connect to the MCP service at a fixed endpoint using AWS IAM credentials with SigV4 signing. The MCP service is available when agent access is enabled on a stack. +