MCP vs A2A
Last verified 2026-06-10Short answer: MCP and A2A are complementary layers, not competitors. MCP, the Model Context Protocol (Anthropic, 25 November 2024, now a Linux Foundation project), connects a single AI agent to external tools, data and business apps, the agent-to-tool layer. A2A, the Agent2Agent Protocol (Google, 9 April 2025, donated to the Linux Foundation on 23 June 2025), connects different agents to each other so they can discover, delegate and coordinate, the agent-to-agent layer. A common phrasing: MCP connects an agent to its tools, A2A connects agents to one another. In the commerce stack, AP2 is built as an extension of A2A, and both standards now sit under the Linux Foundation.
| Criterion | MCP | A2A |
|---|---|---|
| Author / steward | Anthropic, now Linux Foundation (AAIF) | Google, now a Linux Foundation project |
| Announced | 25 Nov 2024 | 9 Apr 2025 |
| Open-source status | Open source, MIT | Open source, Apache 2.0 |
| What it is | Agent-to-tool / data context layer | Agent-to-agent communication protocol |
| Scope | Connect agents to tools, data and business apps | Let agents discover each other, delegate tasks and coordinate |
| Payment type | None (not a payment protocol) | None (AP2 extends A2A; crypto via the A2A x402 extension) |
| Transport | JSON-RPC 2.0 (stdio, streamable HTTP) | HTTP-based; spec v1.0 and SDKs at a2a-protocol.org |
| Governance | Agentic AI Foundation (Linux Foundation), donated 9 Dec 2025 | Linux Foundation Agent2Agent project (23 Jun 2025) |
| Notable adoption | Claude, Cursor, Copilot, Gemini, VS Code, ChatGPT | AWS, Cisco, Google, Microsoft, Salesforce, SAP, ServiceNow |
Table last verified .