2.3.2. MCP and A2A for Agent Interoperability
💡 First Principle: Two standards solve two different connection problems. MCP (Model Context Protocol) standardizes how an agent connects to tools and data sources. A2A (Agent-to-Agent) standardizes how agents talk to each other. They sit at different layers and are complementary, not competing.
Why care: the exam was refreshed specifically to include agentic interop (MCP, A2A, Foundry). MCP lets an agent consume tools from any MCP-compliant server without bespoke integration — Foundry's Hosted agents reach managed tools through a Toolbox MCP endpoint. A2A lets agents built by different teams or frameworks coordinate. Knowing which standard addresses tool-connection versus agent-communication is a clean, testable distinction.
| Standard | Connects | Layer | Analogy |
|---|---|---|---|
| MCP | Agent ↔ tools/data | Tool integration | A universal plug for tools |
| A2A | Agent ↔ agent | Inter-agent comms | A shared language between workers |
🧪 Preview status (verified June 2026): Connecting an agent to an MCP server (the MCP tool) is now GA, but Toolbox / Toolbox MCP endpoints remain in public preview, as do A2A and several hosted-agent capabilities. Expect the conceptual distinction (MCP = tools, A2A = agents) to be stable even as the specific preview APIs shift.
⚠️ Exam Trap: "Choose MCP or A2A for your agent system." They're not mutually exclusive — a system can use MCP to attach tools and A2A to let agents collaborate. Treating them as an either/or is the trap baked into the misconception.
Reflection Question: A vendor's agent (built on a different framework) must both use your internal search tool and coordinate with your triage agent. Which standard covers the tool access, and which covers the coordination?
