Why do AI agents remain fragile and tightly coupled? Most production implementations lack a standardized way to interact with the OS and external tools. This session explores building a modular, cloud-native agent stack using Model Context Protocol (MCP)—an open standard acting as a "USB-C port" for AI interoperability.

In this talk, we’ll share how we integrated MCP into an open AI agent architecture using:
1) vLLM for high-performance model inference
2) LlamaStack as the agentic orchestration framework
3) MCP for standardized tool invocation and data flow
4) Kubernetes for managing distributed agent workloads

We’ll demo the system and share lessons on scaling these components in a cloud-native environment like Kubernetes. You’ll learn how MCP addresses the interoperability gap and helps avoid vendor lock-in. Attendees will walk away with a roadmap for building modular, vendor-neutral AI applications that are firmly in the hands of the community.