a2a-protocol

A2A Protocol (Agent2Agent)

Open protocol for agent-to-agent discovery and communication. Handles agent cards, message routing, task management, and SSE streaming responses between agents. The wiki’s first reference; first appeared via aimock’s A2AMock surface in the AIMock launch source.

Stub — first appearance in the wiki via the AIMock launch. Will grow when a primary source covers A2A directly.

• GitHub: github.com/a2aproject/A2A

What it provides

• Agent card discovery — agents publish a .well-known/agent-card.json describing their identity, skills, and capabilities
• JSON-RPC message routing — structured request/response between agents
• Task management — long-running tasks with incremental status updates and artifact streaming
• SSE streaming — server-sent events for incremental task progress and partial results

Why it’s in the wiki

A2A is the agent-to-agent counterpart to MCP (agent-to-tools) and AG-UI (agent-to-frontend). Together they form the connective layer of the 2026 agentic stack:

Protocol	Connects
MCP	Agents ↔ tools
A2A	Agents ↔ other agents
AG-UI	Agents ↔ frontend applications

A2A is also the architectural successor to the wiki’s existing multi-agent coverage. Where claude-code-subagents and paperclip track multi-agent patterns inside a single tool’s runtime, A2A is the wire protocol for multi-agent coordination across runtimes — a Claude Code agent talking to a LangGraph agent talking to a Mastra agent. The wiki should track A2A adoption as the indicator for whether multi-agent systems become genuinely portable across vendors or stay siloed in each tool’s harness.

Connection to LLM design patterns

A2A is the protocol-layer enabler for pattern-4-agent-teams. Claude Code’s experimental Agent Teams primitive runs inside one Claude Code instance; A2A is the cross-runtime version of the same pattern. If A2A adoption grows, expect agent-team-style coordination to become a generic capability rather than a per-tool feature.

See Also

• aimock — A2AMock is the first wiki-tracked test primitive for A2A
• mcp — sibling protocol (agent ↔ tools)
• ag-ui — sibling protocol (agent ↔ frontend)
• claude-code-subagents — single-runtime multi-agent primitive (compare/contrast)
• llm-design-patterns — Pattern 4 (Agent Teams) is the in-runtime version of what A2A enables across runtimes
• multi-agent-orchestration
• Source: AIMock launch (first A2A reference)