Quarkus Agent MCP com Claude Code, Codex e Cursor: a decisão arquitetural que passou despercebida

Did your AI agent crash along with the Quarkus application you were building? That's the kind of problem the Quarkus Agent MCP solves. In this video, I break down the architectural decision the Quarkus team made when building the Agent MCP: to run as a standalone process, separate from the application you are developing. It seems like a detail, but it changes the workflow for those who use AI agents (Claude Code, Codex, Cursor, IBM Bob, and any other MCP-compatible client) to write Quarkus code on a daily basis. What you will see: Why does Agent MCP exist if Quarkus already has Dev MCP integrated? What changes when the application crashes during development? The three typical usage flows: creating a new project, adding extensions, and inspecting a running application How the Quarkus Skills feature fits into this story Points you need to consider before adopting: another process on the machine, network communication between processes, and the server's broad access level Minimum checklist of prerequisites to start using it today Links mentioned in the video: Official Quarkus Agent MCP post, with installation steps for each MCP client: https://quarkus.io/blog/introducing-a... Project repository on GitHub: https://github.com/quarkusio/quarkus-... Java AI Specialist Method, my program for Java devs who want to integrate Java with AI in the environment Corporate: https://eldermoraes.ai