Google Launches ADK for Go — Build Production-Ready AI Agents Natively in Go

Overview

Google has released the Agent Development Kit (ADK) for Go, an open-source framework that brings agent development into the Go ecosystem. ADK has already supported Python and Java; the Go SDK now lets backend and AI developers write agent logic, orchestration, and tool integrations directly in Go, keeping the entire workflow inside a familiar toolchain and deployment model.

What ADK Brings to Developers

ADK is a code-first framework designed for building, evaluating, and deploying AI agents. While optimized for Gemini and Google Cloud, its architecture is model- and deployment-agnostic. Key features include:

• Agent behavior and orchestration expressed as normal source files
• Support for workflow agents with sequential, parallel, and looped control flows
• A rich tool ecosystem: built-in tools, custom function tools, OpenAPI tools, Google Cloud tools, and ecosystem tools
• Deployment options including local execution, containers, Cloud Run, and Vertex AI Agent Engine
• Built-in evaluation and safety patterns integrated with Vertex AI Agent Builder

For practical purposes, ADK treats an agent as a service: you run it locally, inspect traces, and then deploy it to a managed runtime rather than keeping it as an ad-hoc script that calls an LLM.

What ADK for Go Adds

The Go release preserves the core ADK capabilities while exposing them through an idiomatic Go API. Google positions ADK for Go as a performant, strongly typed way to build agents that leverage Go concurrency patterns. Highlights:

• Install with: go get google.golang.org/adk
• Source is open on GitHub at github.com/google/adk-go
• Supports building, evaluating, and deploying sophisticated agents with the same abstractions for agents, tools, and workflows used in other ADK languages

This enables embedding agent behavior inside existing Go services without switching languages. Teams can compose multi-agent systems where each agent is a Go component, interoperating through the ADK framework.

Agent-to-Agent (A2A) Protocol Support

ADK for Go includes native support for the Agent2Agent (A2A) protocol. A2A standardizes how agents call other agents over a secure interface. With A2A in Go, a primary agent can orchestrate and delegate tasks to specialized sub-agents that run locally or remotely. Interactions remain secure and opaque so that private memory or proprietary logic need not be exposed.

Google also contributed an A2A Go SDK to the main A2A project, offering Go developers a protocol-level entry point to interoperate with other runtimes that support A2A.

MCP Toolbox and Database Integration

The Go ADK ships with native integration for MCP Toolbox for Databases, providing out-of-the-box support for 30+ databases. MCP Toolbox is an open source MCP server that handles connection pooling, authentication, and exposes database operations as tools using the Model Context Protocol (MCP).

Within ADK, this means agents:

• Register MCP Toolbox for Databases as an MCP tool provider
• Call database operations through MCP tools instead of crafting raw SQL
• Use a set of safe, predefined actions that the toolbox enforces

This fits ADK's tool model where agents combine built-in, Google Cloud, ecosystem, and MCP tools as needed.

Integration with Vertex AI Agent Builder and Agent Engine

ADK is the primary framework supported by Vertex AI Agent Builder for multi-agent development. The recommended workflow is:

1. Develop and test agents locally using ADK (now with Go support)
2. Use ADK quickstart and dev UI to test agents with multiple tools
3. Deploy agents to Vertex AI Agent Engine as a managed runtime

For Go teams this means a single language can span the full agent lifecycle — local development, testing, and production deployment — without language transitions.

Practical Implications for Teams

ADK for Go narrows the gap between AI agents and Go-based backends. Teams that already build services in Go can now integrate agents directly into their existing codebases and deployment pipelines. The addition of A2A protocol support and MCP Toolbox integration makes it feasible to build secure, interoperable, and production-ready multi-agent architectures that align with Google Cloud tooling for evaluation and observability.

Resources

Check the ADK Go repository and samples on GitHub for technical details, tutorials, and examples. ADK for Go is open source and intended to be used with Vertex AI Agent Builder and Agent Engine when moving to managed production environments.