# A2A

> The leading standard for agent-to-agent coordination: agents publish an Agent Card describing their skills, then negotiate and delegate tasks to one another.

_The Agent Protocol Atlas · /protocols/a2a · [JSON](/api/protocols/a2a) · [all The Agent Protocol Atlas](/protocols)_

- **name:** A2A
- **full_name:** Agent2Agent Protocol
- **layer:** interop
- **creator:** Google (now under the Linux Foundation)
- **status:** production (150+ orgs)
- **source:** A2A project / Linux Foundation announcements, 2025–2026 (org count per project communications).
- **year:** 2025
- **one_liner:** The leading standard for agent-to-agent coordination: agents publish an Agent Card describing their skills, then negotiate and delegate tasks to one another.
- **spec_url:** https://a2a-protocol.org
- **snippet:**

```
GET /.well-known/agent.json → { "name": "...", "skills": [ ... ], "endpoints": { ... } }
```
- **abbreviation:** A2A
- **also_known_as:** Agent2Agent, Agent-to-Agent Protocol
- **canonical_spec_url:** https://a2a-protocol.org
- **entity_uri:** https://en.wikipedia.org/wiki/Agent2Agent
- **taxonomy_layer:** interop
- **sub_layer:** agent-to-agent
- **protocol_type:** transport
- **central_problem:** Lets independent AI agents on different platforms discover each other, advertise skills, and delegate or negotiate tasks across organizational boundaries.
- **maintainer:** A2A project, hosted by the Agentic AI Infrastructure Foundation (AAIF) under the Linux Foundation
- **governance_body:** AAIF
- **license:** Apache-2.0
- **maturity_tag:** production-ready
- **current_spec_version:** v1.0
- **spec_date:** 2026-04
- **launch_date:** 2025
- **last_verified:** 2026-06-15
- **transport:** JSON-RPC / HTTP, with Server-Sent Events for streaming task updates
- **core_mechanism:** Each agent publishes an Agent Card at a well-known URL describing its identity, skills, and endpoints; a client agent reads the card and opens a task with the remote agent, exchanging messages and artifacts over the task lifecycle until completion.
- **discovery_endpoint:** /.well-known/agent.json (Agent Card)
- **settlement_type:** —
- **adoption_metric:** 150+ organizations, landed in major cloud platforms, enterprise production use within its first year (April 2026) (source: https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year)
- **notable_adopters:** {"value":"Google","source":"https://a2a-protocol.org"}, {"value":"Major cloud platforms (per LF April 2026 press)","source":"https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year"}
- **relationships:** {"predicate":"complements","target":"mcp","note":"A2A -complements-> MCP (research §2 seed triple): A2A is agent→agent, MCP is agent→tool."}, {"predicate":"governed_by","target":"aaif","note":"A2A v1.0 housed by AAIF (research §2)."}, {"predicate":"extends","target":"ap2","note":"AP2 is designed as an extension of A2A's task lifecycle for payments."}
- **ideal_use_case:** Multi-agent systems where agents from different vendors/teams must coordinate and delegate work.
- **when_to_use:** When you need agents to talk to OTHER agents — discover skills, hand off tasks, and stream results across trust boundaries.
- **when_not_to_use:** When a single agent just needs to call tools or read your site (use MCP or Layer-1 discovery files); A2A's agent-card + task machinery is overhead you don't need.
- **code_example:** // Discover a remote agent's capabilities
const card = await fetch('https://remote.example/.well-known/agent.json').then(r => r.json());
// card.skills lists what the agent can do; open a task against card.endpoints
- **agent_readiness_link:** agent-readiness/a2a
- **layer_legacy:** agent