Build Production-Ready Agents
A flexible, modular framework to develop and deploy AI agents that scale. From simple prototypes to enterprise-grade workflows.
pip install google-adk
npm install @google/adk
go get google.golang.org/adk
com.google.adk:google-adk:0.6.0
Next-Gen Developer Tools
Go from idea to agent instantly. Leverage Skills and the Model Context Protocol (MCP) to give your agents powerful, ready-made capabilities.
Your Agent, Your Way
Use natural language to scaffold agents with `claude code`. The ADK skill automatically translates your requirements into robust, tool-equipped agents connected to any MCP server.
from google.adk import Agent, Tool
import requests
# Tool is automatically generated by the Skill
get_weather = Tool(
name="get_weather",
description="Gets weather for a city",
fn=lambda city: requests.get(f"https://api.weather.com?q={city}").json()
)
# Agent configured with the tool
agent = Agent(
name="weather_agent",
model="gemini-2.0-flash",
instruction="You are a helpful weather assistant.",
tools=[get_weather],
)
Visualize, Debug, and Trace
The built-in Dev UI gives you a powerful, interactive web interface for testing and debugging. Chat with your agents, inspect their thoughts, and trace every step of their execution to build with confidence.
Go Beyond Vibes
Reliability is key. ADK's evaluation framework lets you systematically test and measure your agents. Go beyond subjective "vibes" by evaluating not just the final responses, but the entire agent trajectory—thoughts, tool calls, and all.
Running 3 test cases... [PASS] test_weather_agent_sunny: Responded correctly about sunny weather. - Trajectory: thought -> get_weather(city='Palo Alto') -> response [PASS] test_weather_agent_rainy: Identified need for an umbrella. - Trajectory: thought -> get_weather(city='Seattle') -> response [FAIL] test_weather_agent_unknown_city: Failed to handle invalid city. - Trajectory: thought -> get_weather(city='Nowhere') -> ToolError 2 passed, 1 failed.