A2A: Protocol Hub

When agents from different platforms need to work together—discovering capabilities, exchanging messages, coordinating tasks—they must speak a common protocol. When each agent stack uses custom APIs, you spend more time wiring integrations than scaling collaboration. What happens when every AI assistant uses a different handshake? You build one-off bridges instead of interoperable agent networks. This page is for technical leads, solution architects, and backend engineers implementing the Agent-to-Agent (A2A) protocol, the open standard for secure, auditable agent collaboration. It covers A2A's scope, canonical data model (AgentCard, Task, Message, Artifact, Part), abstract operations and transport bindings, discovery and security, and an end-to-end implementation checklist. Example payloads and pitfalls help teams launch faster and stay compliant. The guidance aligns with eLLMo AI's protocol-first, governance-forward approach.

A2A (Agent-to-Agent) is an open protocol enabling independent agents to discover one another, exchange messages, coordinate task lifecycles, and share artifacts securely across frameworks and vendors. It standardizes a canonical data model (AgentCard, Task, Message, Artifact, Part), abstract operations (SendMessage, SubscribeToTask, GetTask), and transport bindings (JSON-RPC/HTTP, REST, SSE/gRPC) to deliver interoperable, auditable collaboration for agentic commerce.

Source: A2A Core Specification

A2A v1.0 organizes concerns into three layers: canonical data model, abstract operations, and protocol bindings.

Source: A2A Core Specification

A2A architecture follows a three-layer model that separates data structures, operations, and transport mechanisms.

A2A coordinates with MCP (capability discovery), UCP (product/offer distribution), and ACP (commerce flows). The diagram below shows how these protocols work together in agentic commerce.

Use explicit headers for compatibility:

• A2A-Version: e.g., 1.0
• A2A-Extensions: comma-separated extension identifiers (if any)

Enforce strict schema validation for forward/backward compatibility.

Source: A2A Core Specification

curl -H "Accept: application/json" https://brand.example/.well-known/agent-card.json

{
  "name": "BrandCommerceAgent",
  "description": "Order, inventory, and policy-aware commerce agent",
  "url": "https://api.brand.example/a2a",
  "version": "1.0.0",
  "capabilities": {
    "streaming": true,
    "pushNotifications": true
  },
  "endpoints": {
    "rpc": "https://api.brand.example/a2a/rpc",
    "events": "https://api.brand.example/a2a/events"
  },
  "security": {
    "auth": ["oauth2-jwt", "mtls"],
    "did": "did:web:brand.example"
  }
}

Source: A2A Discovery; Data Structures

POST /a2a/rpc HTTP/1.1
Host: api.brand.example
Content-Type: application/json
Authorization: Bearer eyJhbGciOi...
A2A-Version: 1.0

{
  "jsonrpc": "2.0",
  "method": "SendMessage",
  "params": {
    "message": {
      "role": "user",
      "parts": [
        {"type": "text", "text": "Find in-stock SPF 50 sunscreen under $40 and start checkout."}
      ],
      "contextId": "conv-9c2d",
      "metadata": {
        "locale": "en-US",
        "currency": "USD"
      }
    }
  },
  "id": "req-001"
}

{
  "jsonrpc": "2.0",
  "result": {
    "message": {
      "role": "agent",
      "parts": [
        {"type": "text", "text": "I found 3 products. Would you like me to add #2 to cart?"}
      ]
    },
    "task": {
      "id": "task-7f88",
      "status": {"state": "working", "message": "Orchestrating checkout"}
    }
  },
  "id": "req-001"
}

GET /a2a/events?taskId=task-7f88 HTTP/1.1
Host: api.brand.example
Accept: text/event-stream
Authorization: Bearer eyJhbGciOi...
A2A-Version: 1.0

Event stream (server → client)

event: task.update
data: {"task":{"id":"task-7f88","status":{"state":"working","message":"Inventory reserved"}}}

event: task.update
data: {"task":{"id":"task-7f88","status":{"state":"completed","message":"Order #17293 placed"}}}

{
  "jsonrpc": "2.0",
  "method": "CancelTask",
  "params": {"taskId": "task-7f88"},
  "id": "req-009"
}

# Pseudocode: JSON-RPC handler
def handle_rpc(req):
    method = req["method"]; params = req.get("params", {})
    if method == "SendMessage":
        # Validate auth, version, schema
        # Route to executor; optionally create task
        return {"message": {"role": "agent","parts":[{"type":"text","text":"Ack"}]}}
    if method == "GetTask":
        # Return durable task state
        return {"task": lookup_task(params["taskId"])}
    if method == "CancelTask":
        # Idempotently cancel and emit event
        return {"task": cancel_task(params["taskId"])}
    raise MethodNotFound()

eLLMo connects your existing catalog and business policies to AI agents without rebuilding your tech stack. We organize and verify your product information so every AI-driven interaction reflects your real inventory, pricing, and brand rules. When customers ask AI assistants to find, recommend, or purchase your products, they get accurate answers and your business gets completed orders. You gain complete visibility into how agents are discovering and converting your customers.

See: Agentic Commerce; Solutions for Brands

Protocols hub (parent); Integrations hub (stack-first guides); Glossary: Agentic commerce; Blog: API-first vs protocol-first; Blog: Agentic Commerce Protocol Landscape 2025-2026.