Demo: AI agents talking to each other
A Research agent asks a Writer agent for a summary. Press Run to play the whole exchange, or Step to advance one message at a time. Every message is shown in full below.
Agent A: Research (A2A client)
- State
- idle
- Doing
- Waiting to start
- Has
- An MCP tool: notes_search
Agent B: Writer (remote agent)
- State
- idle
- Doing
- Waiting to be contacted
- Internals
- Hidden from Agent A (opaque)
Message log
What to notice
- The first two messages are MCP: an agent using a tool. The rest are A2A: an agent working with another agent.
- Agent A reads an Agent Card before sending work.
- The work is a task with states. The Writer pauses in input-required, and Agent A answers on the same task.
- The result is an artifact, not just a chat message.
- Agent A never sees how the Writer works inside.
Compare the two protocols in the A2A vs MCP table.