Building Agents with MCP and Data Streams

LLMs are incredibly powerful, but they have two problems: they only know what they read on the Internet, and they can’t actually do anything—they can only chat with you. If you want to build agentic applications that have access to the immediate, non-public context of your business, and you want your agents to be able take actions in the world, you’ll probably need some help from the Model Context Protocol, or MCP.

And that “business context” increasingly exists in the form of real-time streaming data, often in Kafka topics. Once you’re asking your microservices to interpret natural-language prompts, then deputizing them to take actions on your behalf—this is what an agent is!—you can’t afford for them to be acting on out-of-data context. They need to remain deeply connected to the events that matter to your business.

In this presentation, we’ll get a solid overview of MCP itself, then see you how you can use it to build practical multi-agent architectures powered by real-time, streaming data. We’ll see what’s possible when we stop thinking about AI as an external chatbot and start treating it as part of our streaming architecture. Agents are here, and they are powered by streams.