Google ADK

The Google Agent Development Kit (ADK) is Google's open-source framework for building multi-agent AI applications. Announced at Google I/O and open-sourced in April 2025, ADK supports hierarchical teams of agents — lead agents delegating to subagents — with shared state, code-native definitions, MCP support, live multimodal interaction, and A2A protocol interoperability across frameworks.

Key Features

• Hierarchical Multi-Agent Design: Lead agents delegate to subagents in tree structures; shared state flows across the team.
• Code-Native: Agents are defined directly in Python (or Java/Go) — no DSL or visual builder required.
• Tool Building: The @tool decorator turns Python functions into agent tools with structured schemas.
• MCP Support: Works as both MCP client and server for tool and resource sharing.
• Live Multimodal: Voice and video interaction support for Gemini live capabilities.
• A2A Interoperability: The Agent2Agent protocol lets ADK agents collaborate with agents from other frameworks and vendors.
• Shared State & Memory: Persistent state management across agents and sessions.
• Gemini Ecosystem: Deep integration with the Gemini Developer API, Vertex AI, and Gemini Enterprise; works with other models via LiteLLM.

Use Cases

Who Should Use This Tool?

• Google Cloud Teams: Organizations on Vertex AI or Gemini Enterprise building production agents.
• Multi-Agent Builders: Teams needing hierarchical agent structures with delegation.
• Cross-Framework Integrators: Companies standardizing on A2A for agent interoperability.

Problems It Solves

1. Agent team coordination: Hierarchical delegation scales beyond flat agent lists.
2. Framework silos: A2A support enables collaboration with agents from other vendors.
3. Multimodal needs: Live voice/video interaction built in for Gemini models.

Pricing

Advantages & Unique Selling Points

Compared to Competitors:

1. First-party Google integration: Deep Gemini, Vertex AI, and Gemini Enterprise ties.
2. A2A-native: The reference framework for the Agent2Agent interoperability protocol.
3. Multimodal first: Live voice/video support that most frameworks lack.

What Makes It Stand Out:

• Code-native agent definitions — no learning a DSL.
• Three languages: Python, Java, and Go.
• Used by Google Cloud's enterprise agent ecosystem.

Getting Started

Quick Start Guide

1. Install: pip install google-adk.
2. Define an agent: Create an Agent with a model, instructions, and tools.
3. Add tools: Decorate functions with @tool.
4. Run locally: Use the ADK CLI or a local runner to chat with your agent.

Integration

Integrates with:

• Gemini Developer API, Vertex AI, Gemini Enterprise
• LiteLLM for other model providers
• MCP servers and A2A agents

Frequently Asked Questions

Do I need to be on Google Cloud to use it?

No — ADK is open source and works with the Gemini API or other models via LiteLLM, but it shines on Google Cloud.

What languages are supported?

Python, Java, and Go.

How does A2A help me?

A2A lets your ADK agents discover and collaborate with agents built on other frameworks or vendors.

Alternatives

If Google ADK isn't the right fit, consider these alternatives:

• LangGraph: Low-level stateful orchestration with durable execution.
• CrewAI: Role-based multi-agent teams.
• OpenAI Agents SDK: OpenAI's minimal production framework.

Tips & Best Practices

1. Use hierarchical teams for scale: Lead agents delegating to specialists handle complex tasks better.
2. Leverage A2A for cross-vendor work: Standardize on A2A when agents span frameworks.
3. Stay code-native: Keep agent definitions in code for versioning and testing.

Conclusion

Google ADK is the framework for building hierarchical multi-agent systems with deep Gemini integration, MCP support, and A2A interoperability — all in code-native Python, Java, or Go. If you're on Google Cloud or standardizing on A2A, ADK is the reference choice.