MCP in the Wild: Real Time Agentic Systems with Live AB Testing

Ever wanted AI agents that think, plan, and chat with each other in real time? Arun Gupta’s session introduces the Multi-Context Protocol (MCP), which splits agent logic into bite-sized modules (planning, reasoning, execution) and hooks them up using Python, CrewAI, and FastAPI. Best part: you can hot-swap LLMs on the fly—GPT-4, Claude, Mixtral—no code rewrites needed.

Then watch it in action with live A/B testing: swap models mid-game, track planning time, replanning frequency, context-switch impact and more. This isn’t just about games—any multi-agent system gets cleaner code, sharper benchmarks, and a protocol-based boost that stomps spaghetti orchestration.

Watch on YouTube