Streamlining Agent-to-Agent Collaboration with Microsoft Foundry

Building effective collaboration between agents just got a whole lot simpler thanks to Microsoft Foundry. The newly launched A2A Tool and the incoming A2A endpoints now support the widely available A2A protocol version 1.0. For those of you using the earlier a2a_preview tool type and protocol version 0.3, rest assured these will remain accessible in preview mode for your existing integrations. Additionally, Hosted Agents can easily incorporate A2A tools via Foundry Toolboxes, which are made available through the MCP.

If you’ve kept up with prior previews, you’ll know that A2A Endpoints were once referred to as the A2A API head. Here’s a handy breakdown:

• A2A Endpoint: This exposes a Foundry agent so it can be discovered and invoked by an external agent.
• A2A Tool: This allows a Foundry agent to invoke another compatible A2A agent.
• Hosted Agents: These utilize the A2A Tool through an exposed Foundry Toolbox as an MCP endpoint.

These features collectively empower you to establish multi-agent systems, where agents can specialise in different tasks, share their skills, and work together securely across various service boundaries.

Why A2A Protocol Matters

Previously, many multi-agent patterns required custom APIs, bespoke adapters, or orchestration logic that was tightly linked to specific frameworks. The A2A protocol brings a standardised approach to agent-to-agent communication. The beauty of this is that one agent can request assistance from another without needing to dive into its implementation specifics. By using an agent card, the calling agent can discover the remote agent’s capabilities, assign tasks via the A2A protocol, and receive a response to enrich the ongoing conversation.

Authentication and Security

When it comes to Foundry-hosted A2A endpoints, security is paramount. Both the agent card URLs and protocol endpoints require Microsoft Entra ID authentication, ensuring they aren’t publicly accessible. For instance:

• A support agent could reach out to a billing agent for assistance.
• A research agent might contact a data-analysis specialist.
• An enterprise agent hosted in Foundry could call upon a specialised agent that operates outside of Foundry.
• A Hosted Agent can access a Foundry Toolbox that connects through A2A.

In this scenario, a Foundry agent acting as an A2A endpoint can be discovered by an external agent through its agent card. It can then invoke it using the A2A protocol version 1.0 and receive a response generated by the Foundry tools and data.

Configuration and Access

To let a Foundry agent operate as an A2A endpoint, you’ll need to set up two key components:

1. An agent card, detailing the agent’s capabilities.
2. The A2A protocol on the agent endpoint.

The agent card can be configured directly in the Foundry portal, allowing other agents to call it via A2A endpoints. Here’s an example of what part of the agent’s PATCH request body might look like:

{ 
  "agent_card": { 
    "description": "A specialist agent that answers questions about invoices.", 
    "version": "1.0", 
    "skills": [ 
      { 
        "id": "invoice-lookup", 
        "name": "Invoice lookup", 
        "description": "Finds and summarizes invoice status." 
      }
    ]
  }, 
  "agent_endpoint": { 
    "protocol_configuration": { 
      "responses": {}, 
      "a2a": {}
    } 
  } 
}

Targeting the A2A protocol version 1.0 is advised for new integrations, with version 0.3 still available for existing ones but not recommended for new builds.

Frequently Asked Questions

What is the A2A protocol?

The A2A protocol is a standardised method for allowing agents to communicate and invoke each other without needing to know implementation details.

How do I set up an A2A connection?

You first need to establish your Foundry project and then create a RemoteA2A connection, specifying the target URL and authentication type.

What are agent cards and why are they important?

Agent cards describe an agent’s capabilities, making it easier for other agents to discover and invoke them through the A2A protocol.

How is authentication handled in A2A connections?

Authentication for incoming A2A endpoints requires Microsoft Entra ID, ensuring a secure connection where the caller must have the necessary permissions.

In summary, with the A2A Endpoints and A2A Tool, Foundry agents become modular, enabling cleaner design and secure collaboration. Need more guidance? Dive into our resources or tutorials to make the most of these powerful new features.