5 Integration Patterns for Multi-Agent Systems from Google Cloud: A2A and MCP

Google Cloud unveiled infrastructure for multi-agent systems at Cloud Next 26. The core is A2A (Agent-to-Agent) for agent-to-agent communication and MCP (Model Context Protocol) for tool and data access.

There are five key integration patterns.

• Agent discovery and registration: An Agent Card exposes capabilities, authentication requirements, and call limits, while an Agent Registry provides centralized search and access to agents within an organization.
• Cross-team delegation: A coordinating agent can delegate tasks to agents from different languages and teams, such as Go, Java, and TypeScript.
• MCP-based tool connectivity: REST APIs, databases, and legacy systems are unified under a single standard interface, and Apigee API Hub can turn existing APIs into tools for agents.
• Cross-organization collaboration: The Agent Gallery lets you use verified partner agents from Adobe, ServiceNow, Salesforce, and others, while Agent Gateway policies control data sharing and the scope of actions.
• Event-driven agent mesh: Agents connected to BigQuery tables or Pub/Sub streams detect events and escalate to specialized agents or humans.

The article views A2A's openness and the division of roles between A2A and MCP as strengths, but notes that dependency on the Google Cloud ecosystem and enterprise operational complexity still need to be verified.