a2a_client_server/

A2A Policy Client/Server (End-to-End)

This sample ports the .NET samples/05-end-to-end/A2AClientServer policy-agent scenario to Go.

It demonstrates:

1. Hosting a policy agent through both A2A HTTP+JSON and JSON-RPC.
2. Discovering that agent and sending interactive requests in a shared A2A session.

Prerequisites

• Go 1.26+
• Microsoft Foundry project environment and authentication for the server:

$env:FOUNDRY_PROJECT_ENDPOINT="https://<your-foundry-service>.services.ai.azure.com/api/projects/<your-project>"
$env:FOUNDRY_MODEL="gpt-5.4-mini" # optional, defaults to gpt-5.4-mini
az login

The client connects directly to the A2A server and does not require Foundry credentials.

Run server

cd examples/05-end-to-end/a2a_client_server/a2a_server

go run . --port 5000

The server listens on localhost and exposes:

• JSON-RPC A2A endpoint at /
• HTTP+JSON A2A operations such as /message:send
• agent card at /.well-known/agent-card.json

A2A_AGENT_URL optionally overrides the URL published in the card. When using a different port, set the client URL to match it.

Run client

cd examples/05-end-to-end/a2a_client_server/a2a_client
$env:A2A_AGENT_URL="http://localhost:5000/"
go run .

Ask questions like:

• What is the policy for short shipments?
• When should the customer be notified?

The server explicitly uses an in-memory session store for multi-turn requests. This local sample does not authenticate callers. Before exposing it to untrusted clients, authenticate requests and isolate session and task storage by caller; context and task IDs are not authorization boundaries.

Related examples

Set A2A_AGENT_HOST to a trusted server URL for the examples under examples/02-agents/a2a and examples/02-agents/providers/a2a.

• as_function_tools wraps the complete remote agent as one function tool for a Foundry-backed host agent.
• protocol_selection demonstrates an explicit HTTP+JSON or JSON-RPC preference.
• polling_for_task_completion requires a task-capable remote agent and polls its continuation token.
• stream_reconnection interrupts after the first update and reconnects using its continuation token; the remote agent must support streaming tasks.

The policy server uses the default message mode, so it does not issue task continuation tokens. The task-oriented examples require a server configured with a2aprovider.ReturnTask() rather than ReturnMessage(), with streaming advertised for reconnection.