Building and Deploying AI Agents in Minutes with Google’s ADK — Part 1

3 min readApr 11, 2025

Introduction to ADK

The Agent Development Kit (ADK) is Google’s open-source framework designed to simplify the development, deployment, and management of AI agent applications. ADK provides a complete toolkit for building, testing, and deploying multi-agent applications with minimal code.

Getting Started with ADK

Installation

pip install google-adk

Creating a Simple Agent

ADK uses a convention-based approach that dramatically simplifies agent creation. Here’s all you need for a basic weather agent:

Create your agent directory structure:

weather_agent/
  ├── __init__.py
  ├── agent.py
  └── .env

In __init__.py, simply import your agent module:

from . import agent

In agent.py, define your agent and tools:

from google.adk.agents import Agent

def get_weather(location: str) -> str:
    """Get the weather for a specific location"""
    return f"The weather in {location} is sunny with a temperature of 70 degrees."

# Root agent is automatically detected by ADK CLI
root_agent = Agent(
    name="weather_agent",
    description="You are a weather agent that can get the weather for a specific location.",
    tools=[get_weather],
    model="gemini-2.0-flash-exp",
)

In .env, configure your API access:

GOOGLE_GENAI_USE_VERTEXAI="False"
GOOGLE_API_KEY="your_api_key_here"

Running Your Agent

With ADK CLI, running your agent is as simple as:

# Run in interactive CLI mode
adk run weather_agent

# Run with web UI
adk web weather_agent

Example Interaction

When you run the agent in CLI mode, you’ll see an interaction like this:

Running agent weather_agent, type exit to exit.
user: What's the weather in New York?
[weather_agent]: The weather in New York is sunny with a temperature of 70 degrees.

user: How about in Tokyo?
[weather_agent]: The weather in Tokyo is sunny with a temperature of 70 degrees.user: exit

The web UI provides a similar chat experience through a browser interface at http://localhost:8000:

Key Concepts

Agents

Agents are the core building blocks in ADK:

Define capabilities and behavior
Connect to foundation models (like Gemini)
Expose tools for performing actions

Tools

Functions that agents can call:

def search_database(query: str) -> list:
    """Search the database for information"""
    # Database logic here
    return results

Advanced Tools

ADK supports rich parameter types using Pydantic:

from pydantic import BaseModel, Field

class CustomerInfo(BaseModel):
    name: str = Field(description="Customer's full name")
    email: str = Field(description="Customer's email address")

def register_customer(customer: CustomerInfo) -> str:
    return f"Customer {customer.name} registered successfully"

Building Multi-Agent Systems

ADK excels at creating systems of specialized agents:

# In research_agent/agent.py
root_agent = Agent(
    name="research_agent",
    description="Finds relevant information from sources",
    tools=[search_web],
    model="gemini-2.0-pro",
)

# In writing_agent/agent.py
root_agent = Agent(
    name="writing_agent",
    description="Creates well-written content",
    tools=[format_text],
    model="gemini-2.0-pro",
)

Deploying to Google Cloud Run

# Authenticate with Google Cloud
gcloud auth login

# Deploy with web UI
adk deploy cloud_run --with_ui weather_agent

# Make service public
gcloud run services add-iam-policy-binding SERVICE_NAME \
  --region=REGION --member="allUsers" --role="roles/run.invoker"

ADK handles all the complex deployment tasks:

Generating deployment files
Building and pushing Docker containers
Setting up Cloud Run services
Providing a URL for accessing your agent

Deployment Output Example

Building using Dockerfile and deploying container to Cloud Run service [adk-default-service-name]...
...
✓ Deployed to https://adk-default-service-name-abc123456.region.run.app
  
  This service is not public. To make it public, run:
  gcloud run services add-iam-policy-binding adk-default-service-name --region=region --member="allUsers" --role="roles/run.invoker"The agent is now deployed and accessible at: https://adk-default-service-name-abc123456.region.run.app

Your agent is now available on the web and can be integrated with other applications via API calls.

Coming Next

In Part 2 of this tutorial series, we’ll explore how ADK integrates with the MCP (Model Context Protocol)

Part 3 will focus on Agent-to-Agent (A2A) communication, allowing your ADK agents to collaborate with agents from other systems through the open A2A protocol.

Resources: