Scale Forem

Scale YouTube
Scale YouTube

Posted on

Devoxx: MCP in the Wild: Real Time Agentic Systems with Live AB Testing by Arun Gupta

MCP lets you build real-time multiplayer “games” where modular AI agents handle planning, reasoning, and execution—then chat with each other in a neat protocol instead of spaghetti code. Using Python, CrewAI, and FastAPI, you can hot-swap LLMs (GPT-4, Claude, Mixtral, you name it) on the fly and instantly see how each model changes agent behavior.

On top of that, you get live A/B testing built in: swap models mid-game, track metrics like planning time, replanning frequency, and context-switch impacts, then benchmark beyond just token counts or latency. The open-source repo walkthrough shows how this protocol-first design scales to any multi-agent system—ideal if you’re tired of wrestling with brittle orchestration.

Watch on YouTube

Top comments (0)