MCP connects agents to tools and context; ACP carries a coding agent's session to a client; A2A defines tasks and artifacts between independent agent services. None chooses the team's work policy. I would add A2A only when independently deployed agents need a shared interface.

Protocols sit at different edges

MCP: a host or agent client discovers and invokes tools, resources, and prompts from servers. Its optional Tasks extension lets a long-running request return a durable task handle and be polled, but both ends must support it; cancellation is cooperative. It is not the internal team scheduler. MCP architecture, Tasks extension.

ACP (Agent Client Protocol): a client can create or load a coding agent session, send a prompt, receive updates, and request cancellation. This is relevant to the Operator's existing Buzz ACP harness. Its session cancellation does not itself prove that a remote sandbox or process has been killed. ACP overview.

A2A: an independent agent service exposes a task/message/artifact lifecycle, discovery, and status retrieval or streaming. It can be an adapter at a boundary if teams span independently owned services. Internal calls within one trusted controller need not adopt it immediately. A2A specification.

This mapping is a proposed interface choice for this architecture, not a decision to implement all three protocols.