From 7d612a24a61d600af8002575125e24b1169240da Mon Sep 17 00:00:00 2001 From: rkoster Date: Thu, 2 Jul 2026 12:02:09 +0200 Subject: [PATCH] Add research note: A2A Protocol (agent-to-agent interoperability) --- research/a2a-protocol.md | 47 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 47 insertions(+) create mode 100644 research/a2a-protocol.md diff --git a/research/a2a-protocol.md b/research/a2a-protocol.md new file mode 100644 index 0000000..603acf0 --- /dev/null +++ b/research/a2a-protocol.md @@ -0,0 +1,47 @@ +--- +title: "Agent2Agent (A2A) Protocol — Agent-to-Agent Interoperability" +author: Ruben Koster (@rkoster) +date: 2026-07-02 +tags: [inter-agent-comms, ecosystem-survey] +cf_areas: [capi, uaa] +status: draft +sources: + - https://a2a-protocol.org/latest/ + - https://github.com/a2aproject/A2A +--- + +## Summary + +Agent2Agent (A2A) is a Linux Foundation open standard (originated at Google) for AI agent +communication. It defines Agent Cards for capability discovery and Tasks as the core +delegation primitive, with no central registry required. It is designed to complement MCP: +MCP connects agents to tools, A2A connects agents to agents. + +## Key findings + +- **Agent Card**: Self-discoverable JSON document describing an agent's capabilities, skills, + supported formats, and endpoints — how agents advertise themselves to the world. +- **Tasks**: Core communication abstraction with `id`, `status`, `history`, and `artifacts`. + Supports streaming and async operations. +- **Decentralized discovery**: Client agents discover remote agents via Agent Card — no + central registry required. +- **Enterprise features**: Built-in authentication, authorization, task delegation with + context preservation, and an extensions mechanism for custom behavior. +- **MCP/A2A complementary**: MCP handles agent-to-tool communication; A2A handles + agent-to-agent coordination. Both are needed for a full multi-agent system. + +## CF relevance + +If agents are first-class workloads on a platform, they need a standard way to discover and +call each other — analogous to how HTTP serves traditional web services. A2A's Agent Card +model raises questions about where capability discovery lives on a platform: route metadata, +a service registry, or a dedicated agent registry. A2A's task delegation model also implies +stateful, potentially long-running calls, which intersects with how platforms handle async +work and routing. + +## Open questions + +- How do agents discover each other within a bounded deployment (e.g., an org or space)? +- Should A2A be a natively supported protocol alongside HTTP, or an application-level concern? +- How to handle authentication between agents with different trust boundaries? +- What's the right scope for an "agent registry" — per-space, per-org, global?