A2A PROTOCOL MESSAGING & DISCOVERY

Project Objective

This project demonstrates a multi-agent system using the Agent-to-Agent (A2A) protocol with dynamic discovery via a central rendezvous remote service hosted on cloudflare workers.

The primary goal is to showcase how autonomous agents can dynamically discover, connect, and coordinate with each other to solve complex user requests without hardcoded dependencies.

The system is composed of 3 agents: Personal Assistant, Travel Agent, and Airline Agent with their distinct capabilities. Only the AA can book flights, the PA holds user passport number and the TA can coordinate travel arrangements. For demo purpose, no agent available to book a hotel or rentals.

Each agent registers its A2A agentcard to a central rendezvous service, cornerstone to the dynamic discovery: An agent will query the rendezvous service to find peers with required capabilities and their A2A service endpoint (how to connect).

Communication is standardized using A2A protocol, enabling in particular interoperability between agents with different programming languages and frameworks and long running tasks execution.

Event Flow Window

All communication events are logged in the Event Flow Observatory to provide a transparent view of the communication between agents.

-> See every `discovery`, `handshake`, `call`, and `response` Events as it happens.

• DISCOVERY: Agent searching registry.
  Output: Details of agents capable of performing the task.
• HANDSHAKE: Verification of a peer's status.
  Output: Up-to-date agentcard confirming alive status and capabilities.
• CALL: Task delegation between agents.
  Output: Message sent from Requester to Responder.
• RESPONSE: Result of the task delegation.
  Output: Message sent from Responder to Requester.

How it works

• 1. User Request: You ask the Personal Assistant (PA) to book a trip.
• 2. Discovery: Agents query the Rendezvous Registry to find peers with required capabilities (e.g., "plan travel", "book flight", "get passport").
• 3. Handshake: Agents verify each other and exchange "agent cards" to confirm capabilities.
• 4. Delegation: Tasks are delegated down the chain (PA -> Travel Agent -> Airline).
• 5. Resolution: Results propagate back up to the user.

Agents