Empowering AI Agents with MCP and A2A

As AI agents continue to mature in 2025, the next challenge is making sure they can seamlessly communicate—not only with tools and models, but also with each other. This talk, “Empowering AI Agents with MCP and A2A,” delivered by Dr. Bin Liu, offered a deep dive into exactly how to achieve this.

🚀 What’s MCP and Why Does It Matter?

The Model Context Protocol (MCP) acts like the “USB-C” for AI agents, enabling standardized, model-agnostic integration of large models (like LLMs, VLMs, MLLMs) with tools, databases, and APIs. Built on a JSON-RPC client-server model, MCP helps reduce brittle custom glue code, making it easier to connect models to data and functions reliably.

🤝 Enter A2A: Agent-to-Agent Protocol

While MCP focuses on vertical integration (models ↔ tools/data), the Agent-to-Agent Protocol (A2A) is all about horizontal collaboration. It standardizes how agents exchange messages, delegate tasks, and build distributed workflows—like microservices, but for intelligent agents. With A2A, specialized agents (for maps, finance, healthcare, etc.) can coordinate effortlessly.

🔍 Putting It All Together

So while MCP handles the vertical connections (models to tools and data), A2A handles the horizontal connections (agent to agent). Both can be combined to build powerful, modular AI ecosystems where different pieces of the puzzle work together cleanly.

🛠 Live Demo: MCP in Action

The session concluded with a live demo showing how to set up a local MCP server to perform CRUD operations on files. It was a simple yet powerful illustration of how these protocols work under the hood—showing developers they can start experimenting without heavy infrastructure.

📥 Want to Dive Deeper?

For those interested, here’s the original presentation PDF for a closer look:

➡️ Download the PDF

If you’re curious about more talks like this, or want to join a community passionate about data, AI, and real-world applications, check out and join our Meetup group:

👉 Data & AI Meetup

‍