Agent Development Kit 101: Building a simple Research Agent | by Nathaly Alarcon Torrico | Google Cloud – Community | Jun, 2025

The Agent Development Kit (ADK) from Google is a versatile framework for building intelligent AI agents capable of tasks like web research. Launched at Google Next 2025 in Las Vegas, ADK supports Python and Java, enabling developers to create customizable agents with seamless integration into Google’s ecosystem. In this tutorial, we’ll explore how to build research agents using Python and Google Colab.

In this tutorial, we’ll build a research agent with web search capabilities in two variants: one using Crew AI’s Serper tool and another using ADK’s built-in Google Search tool.

Let’s begin!

Crew AI is a framework for creating collaborative AI agents, and its Serper tool enables web searches via the Serper API.

ADK’s flexibility allows integration with tools from other agentic frameworks like Crew AI. In this use case, we’ll build a researcher agent that uses the Serper API, a service for retrieving real-time web search results, to perform targeted internet searches.

To follow along, obtain an API key from Google AI Studio and a Serper API key from serper.dev. Then, define the tool as follows:

# Let's define the model to use
LLM_MODEL = "gemini-2.0-flash-lite"

# Ensure SERPER_API_KEY is set in your environment
from crewai_tools import SerperDevTool
SERVER_API_KEY = "" # Add here your Serper API Key
os.environ['SERPER_API_KEY'] = SERVER_API_KEY
serper_tool_instance = SerperDevTool(
n_results=3,
save_file=False,
)

Now, we just instantiate the Tool:

adk_serper_tool = CrewaiTool(
name="InternetNewsSearch",
description="This tool performs targeted searches on the internet to find recent news articles. Use it when you need to gather up-to-date information on current events or specific topics that are likely to be covered in news publications.",
tool=serper_tool_instance
)

Next, define the agent using ADK’s Agent class, specifying its name, model, description, instructions, and tools. The tools parameter links the agent to the Serper tool:

serper_agent = Agent(
name="internet_research_agent", 
model=LLM_MODEL,
description="An expert agent designed to find answers to user questions by performing **targeted searches across the internet** using the Serper API.",
instruction="My primary function is to **accurately answer your questions by searching the internet for relevant information**. Provide clear and concise queries for the best results.",
tools=[adk_serper_tool]
)

The agent is ready! Now we can start interacting with it.

Access the full code in this Colab notebook: https://github.com/nathalyAlarconT/adk-training-notebooks/blob/main/ADK_Basic_Serper_Agent.ipynb . Open it and ensure your API keys are set, and run all cells to test the agent.

Having explored external tool integration, we’ll now use ADK’s built-in Google Search tool, which simplifies setup by eliminating the need for external API keys and also, we don’t need to instantiate the tool.

from google.adk.tools import google_search

search_agent = Agent(
name="Google_Search_Expert_Agent",
model=LLM_MODEL,
description="This agent is specialized in **retrieving highly relevant and up-to-date information directly from Google Search** to answer a wide range of questions. It's built for accuracy and efficiency in web information gathering.",
instruction="I am your go-to expert for finding **accurate and current answers by leveraging the power of Google Search**. Just ask me any question, and I will efficiently scour the web to provide the most relevant information. For best results, articulate your queries clearly.",
tools=[google_search]
)

The built-in Google Search tool often provides more comprehensive results due to its direct integration with Google’s search engine.

Access the full code here: https://github.com/nathalyAlarconT/adk-training-notebooks/blob/main/ADK_Basic_Search_Agent.ipynb Run it in Google Colab with your Google AI Studio credentials to test the agent.

The code within the notebooks is really simple to follow. You’ll see that ADK is a simple-to-use framework for building your AI Agents.

I’ll be posting more ADK examples; this was just a simple introduction. Follow for more!