MCP vs A2A: two layers of one stack, and why orchestrators still do all the real work

Diarnstinc_Crew_6510 · reddit · 2026-09-15

The author maps where MCP ends and A2A begins in coding-agent workflows, concluding they're different layers of one stack rather than competitors: MCP gives an agent standardized access to tools, APIs, files and external resources; A2A lets independent agents discover each other, exchange messages, delegate tasks and report progress. One makes a single agent more capable; the other enables separate agents to work together.

In practice, most multi-agent coding setups remain one orchestrator calling several workers — workers use MCP tools internally but rarely talk to each other, and the orchestrator handles every handoff. That's far from what A2A was designed for, raising doubts about real adoption beyond demos. The author's take: MCP plus a decent orchestrator covers agents within one repository; A2A only matters once agents span different services, frameworks, teams or vendors where no single controlling process can be assumed.

Original post →

More from coding & agent

coding & agent channel →