A2A (Agent-to-Agent)

Vocion agents can participate in agent-to-agent (A2A) systems — discovered and invoked by other agents, not just by humans or app code. Where MCP exposes Vocion's tools to a client, A2A exposes a Vocion agent as a peer that another agent can delegate work to and get a structured, reviewable result back.

When to use A2A

• An external orchestrator (or another team's agent) needs to delegate a task to one of your Vocion agents.
• You're composing a multi-agent system where Vocion owns a specific capability — drafting, triage, research, review — behind a stable contract.
• You want delegated work to keep Vocion's production guarantees: context, human review, observability, evals, and budgets all still apply to an A2A-invoked run.

How it works

• Each exposed agent advertises a capability descriptor — its purpose, inputs, and outputs — so a calling agent can discover and invoke it.
• A call runs through the same operating loop as any other interface: the agent reads its context, runs, and — where the agent's review policy requires it — pauses for human approval before returning.
• The response carries the result plus its trace metadata (context version, workflow version, cost), so the caller can attribute and audit the output.

A2A vs MCP

• MCP — Vocion as a tool server: a client (Claude Code, Cursor, your app) calls Vocion's tools. See MCP Server.
• A2A — Vocion as an agent peer: another agent delegates a whole task to a Vocion agent and receives a reviewed result.

Both run on the same runtime and controls; pick the interface that matches who's calling — a client/tool consumer (MCP) or another agent (A2A).