Topic overview
Learn how agents advertise their capabilities via an agent card, how a client agent opens a task with a remote agent, and how results stream back as artifacts and status updates over a standard HTTP-based transport.
Core concepts
Agent cards (capability discovery), tasks (the unit of work exchanged between agents), artifacts (the output an agent produces), streaming status updates, and opaque agent execution — a caller doesn't need to see a remote agent's internal reasoning or tools, only its inputs and outputs.
Why it matters
MCP solves agent-to-tool integration; it doesn't standardize how two agents built on different frameworks or by different vendors negotiate and delegate work to each other. A2A fills that gap, so a multi-agent system can compose specialist agents from different teams or companies without custom glue code per pair.
Interview relevance
Expect questions on how A2A differs from MCP (agent-to-agent vs. agent-to-tool), why opaque execution matters for trust boundaries between organizations, and how task/artifact modeling handles long-running or asynchronous agent work.