Semantic Kernel A2A protocol and A2A Discovery Service Example

This repository showcases a playful yet practical conversation between two autonomous agents using the a2a‑dotnet SDK and Semantic Kernel. It demonstrates how agents can discover each other, exchange JSON‑RPC messages, and stream results back in real time — all while keeping the code simple and transparent.

✨ What makes it cool?

• Agents locate one another through a lightweight discovery service.
• Messages hop over either a local named pipe or Azure Service Bus.
• Semantic Kernel powers text transformations such as reversing or upper‑casing.
• The sample is small enough to grok in minutes yet flexible enough to extend.

Repository layout

The solution is split into several projects to keep responsibilities clear:

Each project targets .NET 8 and uses the latest a2a-dotnet packages.

Prerequisites

• .NET 8.0 SDK
• Optional: An Azure Service Bus connection string if you want to see the agents communicate through the cloud rather than a named pipe.

Running the demo

Visual Studio, set all prject to run at the same time:

Command line

1. Build the solution

   dotnet build
   

2. Start the discovery service

   Agents query this lightweight API to find where other agents are listening.

   dotnet run --project Agent2AgentProtocol.Discovery.Service
   

3. Start Agent 2 (the responder)

   Agent 2 registers the reverse and upper capabilities and waits for work.

   dotnet run --project Semantic.Kernel.Agent2AgentProtocol.Example.Agent2
   

4. Start Agent 1 (the initiator)

   Agent 1 asks the discovery service for an endpoint that can handle a reverse command and then sends a task message.

   dotnet run --project Semantic.Kernel.Agent2AgentProtocol.Example.Agent1
   

5. Watch the conversation

   The terminal output shows Agent 2 reversing text and streaming the response back to Agent 1. Try editing the request in Agent1.cs to send upper: instead of reverse:.

Switching to Azure Service Bus

Want to see the agents chat over the cloud?

1. Create a Service Bus namespace and queue.
2. Set the environment variable SERVICEBUS_CONNECTIONSTRING with a connection string that has Send and Listen rights.
3. In appsettings.json for both agents, change UseAzureServiceBus to true or modify the TransportOptions in Agent2 to set UseAzure = true.
4. Run the steps above; messages now traverse Service Bus instead of a named pipe.

The transport implementation is abstracted behind IMessagingTransport, so you can substitute other mechanisms (web sockets, HTTP, etc.) with minimal changes.

Architecture overview

Agent 2 exposes skills via Semantic Kernel functions. When it receives a task that begins with reverse: or upper:, it invokes the appropriate function and replies with the result. The agents can easily be extended with additional skills by updating TextProcessingFunction.cs and registering new capabilities.

Extending the sample

• Add new text transformations by creating Semantic Kernel functions and listing them in BuildCapabilitiesCard.
• Replace the transports with your preferred messaging system by implementing IMessagingTransport.
• Integrate a real discovery mechanism or directory service for more sophisticated routing.
• Use the TaskManager provided by a2a-dotnet to track long‑running jobs or to stream incremental progress.

Contributing and feedback

Pull requests and issues are welcome! This repository is intentionally lightweight; ideas for additional skills, transports, or polish are encouraged.

Resources

• a2a‑dotnet SDK
• Semantic Kernel
• Azure Service Bus

License

Licensed under the MIT License.