A guide to converting ADK agents with MCP to the A2A framework

In this hands-on engineering guide published on the Google Cloud Blog, I outline how to break down the boundaries of standalone AI applications by upgrading Google Agent Development Kit (ADK) agents into fully cooperative components using the Agent-to-Agent (A2A) framework.

By wrapping a standalone agent—specifically a MultiURLBrowser initialized with a firecrawl-mcp tool execution set—into the universal A2A protocol, we allow disparate agentic microservices to dynamically discover, communicate, and delegate tasks to one another.

Step-by-Step Refactoring Blueprint:

• Step 1: Define the Core Agent and its MCP Tool (agent.py) The foundation relies on setting up the core execution logic. Inside agent.py, the _build_agent method initializes the foundational ADK LlmAgent (running on gemini-1.5-pro-preview-0514) and links its corresponding MCPToolset to launch and manage the firecrawl-mcp stdio server parameter requirements securely.
• Step 2: Establish a Public Identity (__main__.py) For secondary enterprise networks to seamlessly discover your microservice, it must broadcast its unique capabilities. This is achieved by creating an explicit AgentSkill declaration (specifying identifiers, descriptive schemas, and usage examples like web scraping execution intents) paired with an AgentCard metadata anchor exposing endpoint networking variables.
• Step 3: Implement the A2A Task Manager (task_manager.py) The ultimate operational bridge relies on inheriting the A2A server’s AgentExecutor interface. By overriding the asynchronous execute and cancel method loops, the manager intercepts remote pipeline invocations, wraps runtime state updates via a localized TaskUpdater, and pushes streaming structural outputs back to the orchestration cluster smoothly.