I hosted and presented a session A2A, MCP, and how independently built AI agents can work together at our Nashua Cloud .NET User Group.
The main idea is simple: MCP connects an agent to its tools, while A2A connects agents to each other. A2A gives agents a standard way to discover one another, understand capabilities through Agent Cards, and collaborate without exposing their internal implementation.
I also covered how this fits into Microsoft Agent Framework, including how to expose an agent through A2A, consume a remote agent, and support patterns such as streaming and long-running tasks.
One of the examples in the session shows a support agent working with separate order, policy, and shipping agents. Each agent can be built and managed independently, while A2A provides the common communication layer between them.
If you are already experimenting with MCP or multi-agent systems, A2A is worth understanding. The two protocols are complementary and will likely be used together in many agent architectures.
If you couldn’t make it, here’s everything:
Watch the session: https://youtu.be/fbWP_aiu5Ps
Github: https://github.com/nhcloud/a2a