Today, we are moving beyond standard chatbots. In this hands-on lab, you will build a deterministic, multi-reasoning Agentic AI system using Google's Agent Development Kit (ADK) and the Model Context Protocol (MCP).
Instead of an AI that just generates text, you will build an orchestrator. Our Transit Agent will dynamically fetch live data, read business policies, and execute transactional workflows completely on its own. Finally, we will serve it through a beautiful web UI!
- Google ADK: The "Brain" - Orchestrates the ReAct (Reason + Act) loop using Gemini.
- FastMCP: The "Muscle" - Exposes your backend APIs and logic to the AI.
- Docker: For reliable, instant deployment.
Before we look at the code, we need a workspace. Follow these steps to set up your Google Cloud environment:
Open the Google Cloud Console.
Click the Project Selector dropdown at the top of the page.
Click New Project.
Name your project (e.g., gdg-mumbai-agent) and click Create.
Once created, make sure your new project is selected in the top dropdown.
Make sure that billing is enabled for your Cloud project. Learn how to check if billing is enabled on a project.
Redeem Credits following the setps mentioned in codelab.
We will use Google Cloud Shell as our terminal so we don't have to install anything locally.
Look at the top right corner of the Google Cloud Console.
Click the Activate Cloud Shell icon (it looks like this: >_).
A terminal will open at the bottom of your screen. Click Continue if prompted.
Once connected to Cloud Shell, you check that you're already authenticated and that the project is set to your project ID using the following command:
gcloud auth listRun the following command in Cloud Shell to confirm that the gcloud command knows about your project.
gcloud config list projectIf your project is not set, use the following command to set it:
gcloud config set project <YOUR_PROJECT_ID>To use Google's generative models, we need to turn on the APIs for this project.
In the Google Cloud Console search bar at the top, type "Vertex AI API" and select it.
Click Enable.
Next, search for "Generative Language API" (used for Gemini API keys) and click Enable.
Alertnatively, run this command in cloud shell:
gcloud services enable aiplatform.googleapis.com generativelanguage.googleapis.com(Note: If prompted to authorize Cloud Shell, click Authorize).
Still in your Cloud Shell open, simply paste this command into your terminal:
git clone https://github.com/ruby-verma/transit-agent.git && cd transit-agent You will need a Gemini API key to give your agent a brain.
Get your API key from Google AI Studio.
Once you have your key, choose one of the following methods to add it to your project:
Option A: The Quick CLI Method (Recommended)
Run this single command in your Cloud Shell terminal, making sure to replace the placeholder with your actual copied key:
In your Cloud Shell terminal, copy the example environment file:
echo 'GOOGLE_API_KEY="YOUR_API_KEY"' > .envOption B: The Visual Editor Method
Copy the template file:
cp .env.example .envUse the Cloud Shell Editor's file explorer on the left side of your screen.
Open the .env file, paste your key inside, and save (Ctrl+S).
When you clone this project, your directory will look like this:
transit-agent/
├── .env.example # Your injected API key
├── Dockerfile # Container spec for the Agent & Server
├── docker-compose.yml # Docker multi-container orchestration
├── README.md # This documentation file
├── transit_server.py # FastMCP Data Server code
└── agents/ # The folder ADK Web scans for agents
└── __init__.py # The ADK Agent orchestration code
This runs our FastMCP Server. It does not contain any AI logic. It simply exposes tools and resources. It simply exposes two types of endpoints to the agent:
Tools: Executable Python functions (e.g., fetching a delay, processing a refund).
Resources: Static text or context (e.g., the corporate refund policy).
This file uses Google ADK. It connects to the MCP server, discovers the tools, and uses Gemini to orchestrate a solution. We use the blazing-fast gemini-2.5-flash model for high rate limits, and we name our variable root_agent so the Web UI can auto-discover it. We have enabled debug logging so you can watch the AI "think" step-by-step.
We have containerized the entire architecture so you don't have to worry about Python versions or dependency conflicts.
In your Cloud Shell terminal, build and run the multi-container application:
docker compose up --buildYou will see the containers build and the server start. Keep this terminal running!
Now for the magic! Let's open the built-in ADK Web UI to interact with our agent.
- Look at the top right of your Google Cloud Shell window.
- Click the Web Preview icon (it looks like a small eye or a window with an arrow).
- Click Change port.
- Type in
8000and click Change and Preview. - A new browser tab will open. Ensure
root_agentis selected in the top-left dropdown!
Try these prompts in the chat box:
The agent will just use the check_transit_status tool.
"What is the status of the Vande Bharat express from Mumbai to Ahmedabad?"
The agent will realize it needs to read the policy://refunds resource.
"My flight from Mumbai to Bengaluru is delayed. What is the official compensation policy for delays?"
The agent will string together a Tool call, a Resource read, and a final transactional Tool call.
"I am travelling from Mumbai to Bengaluru today (User ID: 90210). Please check my flight status. If it is delayed, check the refund policy to see what I am eligible for, and if I am eligible for anything, autonomously process that compensation for me."
When you run the Level 3 prompt, look at the right side of your Web UI screen. You will see the agent's internal monologue as it autonomously calls the status tool, reads the policy, and triggers the final refund!
Built for the GDG Cloud Mumbai Community! Happy Building!