What problem does A2A solve that MCP does not, and how do they work together?

Attempt it

Compare MCP and A2A: what distinct problem does each solve, and describe a system architecture where both protocols are active simultaneously.

Free · 2 AI evals / day

Quick reference

TL;DR

MCP connects one agent to its tools and data (vertical); A2A lets agents delegate tasks to other agents (horizontal). They are complementary, and a real system runs both.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Picture a contractor building a house. MCP is the contractor's toolbelt: a standard way to reach for a drill, a saw, or the blueprint file whenever a job needs it. A2A is the phone the contractor uses to call a licensed electrician from another company when the wiring is beyond their skill. The toolbelt is for things the contractor operates directly. The phone is for handing a whole sub-job to a peer professional you do not control and cannot see inside of. You do not pick one. A serious build uses the toolbelt constantly and the phone whenever a specialist is needed. MCP and A2A play the same two roles for AI agents: reach for tools, or call a peer agent.

Key concepts

Agent-to-Agent protocol AI agents Agent cards Task delegation

Quick scan · 2–4 min

Everything important, quickly.

4 min: the two axes (vertical MCP vs horizontal A2A) + opaque peer model + agent cards + long-running tasks + one architecture that runs both.

Side-by-side

Dimension	MCP	A2A
Axis	Vertical: agent to tools and data	Horizontal: agent to agent
Other end is	A passive tool you drive	An opaque peer agent you ask
Core problem	Capability and context connectivity	Cross-boundary task delegation
Discovery	Server advertises tools, resources, prompts	Agent card advertises skills and endpoint
Interaction shape	Request-response tool calls	Long-running tasks with streamed progress
Visibility into peer	Full: host owns the tool catalog	None: remote internals stay hidden

Real-world examples

Real products, models, and research that use this idea.

Google announced A2A in 2025 with partners like Salesforce and SAP, positioning it as complementary to Anthropic's MCP rather than a replacement.
A customer-service agent in 2026 uses MCP to query an internal CRM and order database, then uses A2A to delegate a shipping question to a partner logistics agent.
Agent cards let an A2A client discover a remote agent's skills and endpoint the same way an MCP host enumerates a server's tools at startup.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow does authentication differ between an MCP tool call and an A2A delegation across organizations?

Sign in or go Pro to see the approach for this follow-up.

QWhy does A2A need a long-running task lifecycle when MCP is mostly request-response?

Sign in or go Pro to see the approach for this follow-up.

QWhen would you wrap a remote agent as an MCP tool instead of speaking A2A to it?

Sign in or go Pro to see the approach for this follow-up.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.