The Agent Development Kit (ADK) is a code-first toolkit from Google for building, evaluating and deploying agents. It gives explicit control over the agent flow, tools and state, without hiding the code from the developer.
What it is for
- Multi-step agents with tools, callbacks and session state
- Multi-agent systems with sequential, parallel and graph workflows
- Custom tools, MCP servers and tools from the Google ecosystem
- Agent evaluation with test sets and metrics
- Deployment to managed environments such as Vertex AI Agent Engine and Cloud Run
How it works
- The developer defines agents, tools and the workflow in code
- The local runtime runs and debugs the agent, with a web development interface
- Sessions and state keep context between interactions
- Support for Python, TypeScript, Go, Java and Kotlin
- Integration with Gemini models and other providers
License and maintenance
- Open source under the Apache 2.0 license, maintained by Google
- Active project in 2026, with ADK 2.0 and graph workflow support
- Managed deployment uses Google Cloud services, billed separately
Strengths
- Code-first approach with fine-grained control of flow and tools
- Support for multiple languages beyond Python
- Evaluation and deployment integrated with the Google Cloud ecosystem
Limitations
- A newer project than the most established competitors
- Part of its production value is tied to Google Cloud