Skip to content
Search
AgenticGlossary

A2A Protocol (Agent2Agent)

Google created A2A, and the Linux Foundation launched it as an open project in June 2025. Each agent publishes an Agent Card, a JSON document that describes its identity, capabilities, skills, endpoint and authentication requirements, so another agent can decide whether to delegate work to it. Work moves between agents as tasks with a defined lifecycle, and results come back as messages and artifacts.

A2A and MCP solve different problems. The A2A documentation puts it this way: MCP connects an agent to tools and data, while A2A lets agents work together as agents, across system boundaries. An agent might use MCP to query a database and A2A to hand a sub-task to a partner's scheduling agent. Because another agent reads your Agent Card to decide whether to call you, its descriptions work like a prompt. Qurtoo covers that in your agent card is a prompt.

Example

# A simplified Agent Card (see the A2A specification for the full set of fields)
{
  "name": "Invoice Reconciler",
  "description": "Matches incoming payments to open invoices and reports anything it can't match.",
  "skills": [
    {
      "id": "match-payments",
      "name": "Match payments to invoices",
      "description": "Given a bank export, returns matched pairs and a list of unmatched payments.",
      "examples": ["Reconcile last week's deposits against open invoices"]
    }
  ]
}

When to use it

  • Agents that need to work with agents run by another team or company
  • Delegating to an agent you don't control and can't import as code
  • Publishing your own agent so others can call it

When not to use it

  • Calling a tool or an API; use MCP or plain function calling
  • All your agents live in one codebase, where direct calls or handoffs are simpler