Docs

Reach · 08.10

A2A: Workflows as Agent-to-Agent Skills

Published workflows exposed as A2A skills through the same execution path as MCP.

A2A

The horizontal complement to MCP's vertical surface

Where MCP is agent-to-tool, A2A (Agent2Agent, protocol version 0.3.0) is agent-to-agent: a workspace publishes an Agent Card advertising its published workflows as skills, discoverable at /v1/a2a/agent-card.json, with per-agent cards available too for finer-grained capability discovery. POST /v1/a2a/message:send addresses a specific skill, runs it, waits for it to finish, and returns an A2A Task carrying the workflow's output as an artifact.

Reception reuses MCP's own execution path, on purpose

A2A task reception runs through the exact same runPublishedWorkflow mechanism the MCP surface uses to execute a published workflow — not a parallel implementation that happens to behave similarly. That's a deliberate design choice: two protocol surfaces sharing one execution path structurally cannot drift out of sync with each other the way two independently-maintained implementations eventually would.

Continue