What problem does it solve? Connecting an AI agent to a remote Windows desktop on Amazon WorkSpaces Applications (AppStream 2.0) involves non-obvious details — SigV4 signing, HTTP-header-based connect modes, SAML federation for domain-joined fleets, and screenshot-budget discipline — that agents frequently get wrong, causing failed connections and wasted automation runs. ## Core Features & Use Cases - MCP Connection Setup: Guides SigV4-signed Streamable HTTP connections to the agentaccess-mcp endpoint, including streaming-URL auth for non-domain-joined fleets and SAML assertion via MCP _meta keys for domain-joined fleets. - Desktop Automation Guidance: Documents the computer-use tool set (screenshot, click, type, key, scroll), BLOCKING vs POLLING connect modes, action batching, and screenshot-budget discipline for reliable GUI automation. - Troubleshooting & Lifecycle: Maps exact error strings (dcv session not ready, client_disconnected, 400 signing-region) to causes and fixes, plus session lifecycle, expire-on-delete, and MCP tool forwarding. - Use Case: An agent needs to open Notepad on a remote Windows fleet, type text, and verify the result — this Skill provides the connection code, the correct POLLING header usage, and the batching strategy to do it without burning screenshot tokens. ## Quick Start Connect my agent to the Agent Access MCP server for my AppStream stack in us-east-1 and take a screenshot of the desktop.