View a markdown version of this page

Agent-to-Agent (A2A) - AWS Prescriptive Guidance
Task Delegation WorkflowOrchestration Layer AgentsA2A + MCP IntegrationSecurity & ObservabilityBenefits

Agent-to-Agent (A2A)

The Agent-to-Agent (A2A) Protocol enables external agents to discover and delegate tasks to Orchestration Layer agents (Q&A Agent, Business Elicitation Agent, Ontology Maintenance Agent) using HTTP + JSON-RPC 2.0 + SSE.

Agent Discovery

Each agent publishes an Agent Card (JSON document) advertising capabilities, schemas, and endpoints. External agents discover cards via Agent Registry, DNS-SD, or ./well-known/agent-card.json.

Q&A Agent Card

{
  "agent_id": "urn:aws:bedrock:...:agent/qa-agent",
  "name": "Question & Answer Agent",
  "capabilities": [{
    "capability_id": "semantic_qa",
    "input_schema": { "question": "string", "ontology_id": "string?" },
    "output_schema": { "answer": "string", "evidence": "array", "citations": "array" }
  }],
  "endpoints": { "invoke": "/invoke" },
  "authentication": { 
    "methods": ["IAM_SigV4", "JWT_OIDC"],
    "required_claims": ["sub", "groups", "act"],
    "scope": "semantic-layer:access"
  }
}

Task Delegation Workflow

1. Discovery → External agent retrieves Agent Card

2. Invocation → Sends JSON-RPC request:

{
  "jsonrpc": "2.0",
  "method": "semantic_qa",
  "params": { "question": "What is the average claim amount for diabetes patients in Q4 2025?" }
}

3. Authentication → JWT/OIDC validation (preferred) or IAM SigV4 for AWS-native callers

4. Execution → Agent orchestrates internal services (Intent Classification, VKG, GraphRAG, Reasoning)

5. Response → Returns result with answer, evidence, citations, confidence

6. Async (Optional) → Long-running tasks return task_id and stream progress via SSE

Orchestration Layer Agents

All agents can support A2A delegation:

  • Q&A Agent — Semantic question answering with federated knowledge graphs

  • Business Elicitation Agent — Business rule extraction, requirement validation

  • Ontology Maintenance Agent — Entity resolution, ontology alignment, truth maintenance

A2A + MCP Integration

A2A (agent-to-agent) and MCP (agent-to-service) are complementary:

External Agent → A2A → Q&A Agent → MCP → Semantic Layer Services

Security & Observability

Auth — Auth - JWT/OIDC (preferred), IAM SigV4 (AWS-native callers) | Authz - Group-based authorization via JWT groups claim with semantic-layer:access scope

Observability — OpenTelemetry traces, CloudWatch Logs, AWS X-Ray

Provenance — All responses include source data, reasoning steps, ontology axioms

Benefits

Interoperability — Agents from different vendors collaborate without custom integration

Scalability — Enterprise-wide agent discovery and delegation

Open Standards — Linux Foundation governance ensures long-term compatibility.