Google ADK
Google's open-source Agent Development Kit for Python, TypeScript, Go, Java and Kotlin: graph workflows, multi-agent teams, MCP tools and stable A2A
Overview
Google's Agent Development Kit (ADK) is an open-source, code-first framework for building, testing and deploying AI agents. It first appeared in April 2025 and now ships SDKs for Python, TypeScript, Go, Java and Kotlin under the Apache-2.0 license; the docs moved from google.github.io/adk-docs to adk.dev. Version 2.0 added graph-based workflows next to the sequential, loop and parallel multi-agent patterns, and ADK TypeScript 2.0 is generally available. Agents get sessions, memory, artifacts and context compaction, and can call plain functions, MCP servers or OpenAPI tools. Live and voice agents are supported. The Agents CLI scaffolds, runs, evaluates and deploys agents, and the built-in web UI helps you debug them. Agent-to-agent (A2A) support lets an ADK agent expose itself to or call other agents; the Python v2.14.0 release in November 2026 marked the A2A integration stable and added an agent-card generator. Gemini is the default, but other models can be used. You can deploy to Google's Agent Runtime on its Agent Platform, Cloud Run or GKE, or to any container host. The framework is free; you pay for model calls and cloud hosting. Main alternatives are LangGraph, CrewAI, the OpenAI Agents SDK and Agno.
The Verdict
Who Should Use Google ADK?
Best For
- Teams building on Gemini and Google Cloud
- Multi-agent systems that need A2A between services
- Polyglot teams that want the same framework in Python, Go, Java or TypeScript
Not Ideal For
- Compliance-heavy production systems that need a long track record
- Teams that want to stay fully cloud-neutral
What's Great
- Fast local loop with adk web, adk run and the API server
- Native A2A and MCP support
- Solid foundation for teams already on Google Cloud
Watch Out For
- Docs thin for real-world production cases
- Multi-user support needs custom work; web UI does not fit per-user agents
- Smaller community than LangGraph; pulls you toward Google Cloud
Pricing
View all features & details
Key Features
- Graph workflows (new in 2.0)
- Sequential, loop and parallel multi-agent patterns
- Sessions, memory, artifacts and context compaction
- Function, MCP and OpenAPI tools
- Live and voice agents
- Agents CLI for scaffold, test, evaluate and deploy
- Evaluation and observability
- Stable A2A integration (Python 2.14.0)
Repositories
- google/adk-python (v2.14.0, Nov 2026)
- google/adk-java (v1.14.0, Nov 2026)
- google/adk-js and google/adk-go
How It Compares
| Feature | Google ADK | LangGraph | CrewAI |
|---|---|---|---|
| Languages | Python, TS, Go, Java, Kotlin | Python, TypeScript | Python |
| Style | Agents plus graph workflows | Graph state machines | Role-based crews |
| Maker | LangChain | CrewAI |
Sources: ADK docs · adk-python · Releases · adk-java · adk-go · adk-js · Thoughtworks Radar
Is this your tool? Add a badge