Agent-to-Agent ProtocolA2A
A standard for AI agents from different systems to communicate and delegate tasks to each other.
Agent-to-Agent (A2A) is an open protocol (Google, April 2025) that standardises how AI agents discover, communicate with, and delegate tasks to each other — across different providers, frameworks, and companies.
The problem it solves: Without a standard, a Claude agent can't easily sub-contract a task to a Gemini agent or a custom enterprise agent. Every integration requires bespoke code. A2A provides the universal handshake.
How it works:
- Agents publish an Agent Card (JSON) at a known URL declaring their capabilities, input/output schemas, and authentication requirements
- The requesting agent calls the target agent via a standard HTTP API, passing a task with context
- The target agent processes and returns a result — synchronously or asynchronously
- Agents can chain: A → B → C → back to A
Relationship to MCP: MCP (Model Context Protocol) connects an agent to tools and data sources. A2A connects agents to other agents. They're complementary — an orchestrator agent uses MCP to access tools and A2A to delegate to specialised sub-agents.
Adoption (mid-2026): 50+ technology partners including SAP, ServiceNow, MongoDB. Integrated into Google's Vertex AI Agent Builder.
In plain terms
Like email's SMTP standard — it lets any email client talk to any other email server. A2A lets any AI agent talk to any other AI agent without custom integration code.
Related concepts
Model Context Protocol
A universal standard for AI to connect to any tool or data source.
AI Agents
AI that plans and takes actions without you guiding every step.
Multi-Agent Systems
Multiple AI agents working together, each handling a specialised role in a larger workflow.
API
A standardised way for software to talk to other software.