ADK Advanced

A comprehensive collection of Google Agent Development Kit (Google ADK) implementations
Demonstrating advanced capabilities, deployment strategies, and production-ready use cases for building intelligent AI agents.

Repository Structure

Directory Overview

#	Directory	Description	Subdirectories
1	adk_tools/	Tool implementations	adk_builtin_tool_agent/ adk_custom_tool_agent/
2	adk_mcp/	Model Context Protocol implementations	adk_mcp_github_agent/ adk_mcp_firecrawl_agent/
3	adk_deploy_agent_engine/	Agent Engine deployment	adk_agent_agentengine_demo/
4	adk_deploy_cloudrun/	Cloud Run deployment	adk_agent_cloudrun_demo/
5	adk_deploy_gke/	GKE deployment	adk_agent_gke_demo/
6	adk_deploy_rag_to_agent_engine/	Agent Starter Pack - Remote Template (RAG - Production-ready)	adk_rag_agent_engine_demo/
7	adk_vertex_ai_search_simple/	Vertex AI Search	vertex_search_agent/
8	adk_vertex_ai_ragengine_simple/	Vertex AI RAG Engine	adk_rag_agent/

Getting Started

Each directory contains self-contained Google ADK agent implementations. Navigate to any subdirectory and follow its README for specific instructions.

Prerequisites

Before you begin, ensure you have the following:

Python 3.11 or later
pip or uv package manager for installing packages
Google API key from Google AI Studio (for local development)
Google Cloud Project (for deployment scenarios)

Installation

Install the Google Agent Development Kit using either pip or uv:

Using pip:

pip install google-adk

Using uv (recommended for faster installs):

# Install uv if not already installed
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install google-adk
uv pip install google-adk

Tip: Consider using a virtual environment to isolate dependencies:

# Using venv + pip
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install google-adk

# Or using uv (automatically manages virtual environments)
uv venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
uv pip install google-adk

Environment Variables Configuration

The GOOGLE_GENAI_USE_VERTEXAI environment variable determines which API backend your application uses:

Using `GOOGLE_GENAI_USE_VERTEXAI=False` (Local Development)

Set this to False (or leave it unset) when:

Developing locally for rapid prototyping using the Gemini Developer API
Authenticating with an API Key from Google AI Studio
Not needing enterprise features and preferring simpler setup

Configuration:

GOOGLE_GENAI_USE_VERTEXAI=False
GOOGLE_API_KEY="your-api-key"

Using `GOOGLE_GENAI_USE_VERTEXAI=True` (Deployment)

Set this to True when:

Deploying on Google Cloud platforms (Cloud Run, Agent Engine, or GKE)
Using Application Default Credentials (ADC) for authentication (standard for code running within Google Cloud)
Needing Vertex AI features:
- Managed sessions via VertexAiSessionService
- Grounding with Vertex AI Search or proprietary data
- Integrated security, data privacy, and data sovereignty controls
- Cloud Trace for observability and performance monitoring
- Using Vertex AI Express Mode (requires this variable plus an API key) [Experimental Preview mode not GA]

Configuration:

GOOGLE_GENAI_USE_VERTEXAI=True
GOOGLE_CLOUD_PROJECT="your-project-id"
GOOGLE_CLOUD_LOCATION="us-central1"

Note: When using GOOGLE_GENAI_USE_VERTEXAI=True, the application authenticates using the service account attached to the Cloud Run service or GKE pod via Application Default Credentials. No explicit API keys are needed in your code or environment variables.

Vertex AI Agent Engine Capabilities

Vertex AI Agent Engine provides a robust, managed platform designed for deploying, running, and scaling AI agents with enterprise-grade features.

Managed runtime for agents: Agent Engine automatically handles deployment, scaling, containerization, and infrastructure security, allowing developers to focus on agent logic.
Sessions & memory bank: Features built-in session management to retain context across interactions, and a persistent memory layer for long-term agent state, crucial for conversational AI.
Tool & model Integration: Enables seamless connection of Large Language Models (LLMs) with various tools, such as APIs or custom function calls, facilitating the creation of complex, end-to-end agent workflows with minimal code.
Security, compliance, observability: Offers enterprise-grade controls including VPC Service Controls (VPC-SC), private networking, comprehensive audit logging, and tracing dashboards for safe, compliant, and transparent operations.

Agent Starter Pack (ASP)

The Agent Starter Pack (ASP) bridges the gap between AI agent experimentation and production. It provides a collection of production-ready GenAI Agent Templates built for Google Cloud, designed to accelerate development through pre-built templates and robust infrastructure for deployment, observability, and CI/CD.

Core Purpose: ASP enables rapid generation of generative-AI agents on Google Cloud by providing deployment infrastructure, observability/monitoring scaffolding, CI/CD pipelines, and best-practice project scaffolding.

ASP Architecture

The following diagram illustrates the high-level architecture of the Agent Starter Pack, showing how components interact from user interaction through deployment and observability:

Key Components:

User & Frontend: User interactions through frontend applications
LLMs: Large Language Models from Vertex AI Model Garden and Gemini
LLM Orchestration: Framework selection (Google ADK, LangGraph, CrewAI) for agent orchestration
Deployment Options:
- Vertex AI Agent Engine (managed serving)
- Cloud Run (containerized services)
- FastAPI (API server)
IaC & CI/CD: Infrastructure as Code (Terraform) and CI/CD pipelines (Cloud Build)
Observability: Cloud Trace, OpenTelemetry, and Cloud Logging for monitoring
Data: Vector stores, BigQuery for data storage, and Looker Studio for monitoring dashboards
Evaluation: Vertex AI Evaluation for assessing agent performance

ASP Features

Production-ready foundation: ASP provides templates for deployment, CI/CD pipelines, infrastructure as code (IaC), API serving, and testing.
Observability & monitoring built in: Includes logging, tracing, and dashboards for comprehensive monitoring of agent performance.
Security & compliance scaffolding: Offers Google Cloud Platform (GCP) best practices for security controls, ensuring enterprise-grade security.
Data, UI & storage integrations: Supports connectivity for data sources, vector stores for RAG, UI playgrounds, and storage layers.
Focus on core logic: ASP allows developers to focus on prompts, tools, orchestration, and business logic while the infrastructure is handled automatically.

ASP Usage

The Agent Starter Pack is used in deployment implementations (folders 3, 4, and 5) to:

Prototype to Production: Accelerate deployment of Google ADK or LangChain agents using ready-made templates.
Enterprise Deployment: Deploy agents with built-in security, monitoring, scalability, and CI/CD pipelines.
Multi-Agent / RAG Workflows: Enable tools, retrieval, and orchestration with data-store & UI support.
Cross-Team Collaboration: Separate agent logic from infrastructure so development and operations teams can work in parallel.

Available Templates:

Agent Name	Description	Use Case
`adk_base`	Base ReAct agent using Google's Agent Development Kit	General purpose conversational agent
`adk_a2a_base`	ADK agent with Agent2Agent (A2A) Protocol support	Distributed agent communication and interoperability
`agentic_rag`	RAG agent with production-ready data ingestion pipeline	Document search and question answering
`langgraph_base_react`	Base ReAct agent using LangGraph	Graph-based conversational agent
`crewai_coding_crew`	Multi-agent system implemented with CrewAI	Collaborative coding assistance
`adk_live`	Real-time multimodal RAG agent with Gemini Live API	Audio/video/text chat with knowledge base

For detailed template information, see the Agent Templates Overview.

Quick Start:

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create a new agent project (interactive CLI)
uvx agent-starter-pack create my-awesome-agent

# Navigate to project and get started
cd my-awesome-agent
make install && make playground

Note: The directory name will match the project name you provide in the create command. For example, uvx agent-starter-pack create adk_agent_agentengine_demo creates a directory named adk_agent_agentengine_demo.

Deployment Targets:

Agent Engine (folder 3): Deploy to Vertex AI Agent Engine
Cloud Run (folder 4): Deploy as containerized services
GKE (folder 5): Deploy to Google Kubernetes Engine

For more information, see the Agent Starter Pack Documentation and GitHub Repository.

Implementations

Core Capabilities

These implementations demonstrate fundamental Google ADK capabilities including built-in tools, custom function tools, and Model Context Protocol (MCP) integrations.

1. Tools

Implementations demonstrating different types of tools available in Google ADK.

Implementation	Directory	Description
Built-in Tools	adk_builtin_tool_agent/	Using Google ADK's built-in tools like `google_search`
Function Tools	adk_custom_tool_agent/	Creating custom tools using `FunctionTool`

2. Model Context Protocol

Implementations using Model Context Protocol (MCP) to connect to external services and tools.

Implementation	Directory	Description
GitHub MCP	adk_mcp_github_agent/	Access GitHub via MCP toolset
Firecrawl MCP	adk_mcp_firecrawl_agent/	Web scraping via Firecrawl MCP client

Deployment Implementations

These implementations demonstrate production-ready deployment patterns using Agent Starter Pack (ASP) with full infrastructure, CI/CD pipelines, and testing frameworks.

3. Agent Engine Deployment

Complete implementation for deploying and managing agents on Vertex AI Agent Engine using Agent Starter Pack (ASP).

Implementation	Directory	Description
Agent Engine	adk_agent_agentengine_demo/	Deploy agent to Vertex AI Agent Engine using ASP templates

Note:

This implementation uses Agent Starter Pack to generate production-ready boilerplate with Terraform infrastructure, CI/CD pipelines, and testing frameworks.

Installation happens at the parent directory level (3_adk_deploy_agent_engine/).

Deployment happens from the child directory (adk_agent_agentengine_demo/).

Testing: After deployment, use jupyter notebook notebooks/adk_app_testing.ipynb to validate your deployed agent.

See 3_adk_deploy_agent_engine/README.md for detailed workflow instructions.

4. Cloud Run Deployment

Complete implementation for deploying agents as containerized services on Google Cloud Run using Agent Starter Pack (ASP).

Implementation	Directory	Description
Cloud Run	adk_agent_cloudrun_demo/	Deploy agent to Google Cloud Run using ASP templates

Note: This implementation uses Agent Starter Pack to generate production-ready boilerplate with Docker containerization, Cloud Run deployment, and CI/CD workflows.

5. GKE Deployment

Complete implementation for deploying agents on Google Kubernetes Engine (GKE) for enterprise-scale deployments.

Implementation	Directory	Description
GKE	adk_agent_gke_demo/	Deploy agent to Google Kubernetes Engine with Workload Identity configuration

Agent Starter Pack - Remote Template (RAG - Production-ready)

Remote templates allow you to create production-ready AI agents from Git repositories using Agent Starter Pack. Any Git repository can be used as a template - the system automatically handles fetching, configuration, and generating your complete agent project.

6. RAG Agent Remote Template

A complete RAG (Retrieval-Augmented Generation) agent template that can be used as a remote template with Agent Starter Pack. This template demonstrates how to create a production-ready RAG agent using Vertex AI RAG Engine, deployed to Agent Engine. Includes full deployment infrastructure, Terraform configuration, CI/CD pipelines, and testing frameworks.

Implementation	Directory	Description
RAG Agent Remote Template	adk_rag_agent_engine_demo/	Remote template for creating RAG agents using Vertex AI RAG Engine, deployable to Agent Engine

Using this template:

# Use this repository as a remote template
uvx agent-starter-pack create my-rag-agent -a https://github.com/arjunprabhulal/adk-advanced/tree/main/6_adk_deploy_rag_to_agent_engine/adk_rag_agent_engine_demo

Note: This template is configured with [tool.agent-starter-pack] in pyproject.toml, making it discoverable and usable as a remote template. See Using Remote Templates for more information.

Vertex AI Search (Simple)

Simple agent implementations using Vertex AI Search for intelligent document search and retrieval. Perfect for local development and testing without deployment complexity.

7. Vertex Search Agent

Simple agent implementation using Vertex AI Search for intelligent document search and retrieval.

Implementation	Directory	Description
Vertex AI Search Agent	vertex_search_agent/	Simple agent using Vertex AI Search for local testing

Vertex AI RAG Engine (Simple)

Simple agent implementations using Vertex AI RAG Engine for enhanced context-aware responses. Perfect for local development and testing without deployment complexity.

8. Vertex AI RAG Engine Agent

Simple agent implementation using Vertex AI RAG Engine for enhanced context-aware responses.

Implementation	Directory	Description
Vertex AI RAG Engine Agent	adk_rag_agent/	Simple RAG agent using Vertex AI RAG Engine for local testing

Quick Start

Follow these steps to get started with any implementation:

Choose an implementation from the directories above
Navigate to the appropriate directory:
- For folders 1-2 (Core Capabilities): Navigate to the parent directory (e.g., 1_adk_tools/) for installation
- For folders 3-5 (Deployment Implementations): Follow the parent directory README for installation, then navigate to the child directory for deployment
- For folder 7 (Vertex AI Search - Simple): Follow the parent directory README for installation and local setup
- For folder 8 (Vertex AI RAG Engine - Simple): Follow the parent directory README for installation and local setup
- For folder 6 (Agent Starter Pack - Remote Template): This is a production-ready RAG remote template that can be used with Agent Starter Pack CLI. See the parent directory README for details on using it as a template
Read the README.md in that directory for specific instructions
Set up your .env file:
- For local development: Copy .env.example to .env and set GOOGLE_GENAI_USE_VERTEXAI=False with your GOOGLE_API_KEY
- For deployment: Set GOOGLE_GENAI_USE_VERTEXAI=True with GOOGLE_CLOUD_PROJECT and GOOGLE_CLOUD_LOCATION
Run with adk run <agent-name> or adk web (agent selection happens in the browser)

Note: Demo screenshots are available in the root images/ folder. Check the individual README files for visual demonstrations of each agent's capabilities.

Example Workflow

For Tools and MCP implementations (folders 1-2):

# Navigate to parent directory
cd 1_adk_tools

# Install dependencies (at parent level)
uv pip install -r requirements.txt
# or: pip install -r requirements.txt

# Navigate to specific agent
cd adk_builtin_tool_agent

# Set up environment variables
cp .env.example .env
# Edit .env with your API keys

# Run from parent directory
cd ..
adk web  # Agent selection happens in browser

For Deployment implementations (folders 3-5):

# Navigate to parent directory
cd 3_adk_deploy_agent_engine

# Install dependencies (at parent level)
uv pip install -r requirements.txt

# Navigate to agent project
cd adk_agent_agentengine_demo

# Set up environment variables
cp .env.example .env
# Edit .env with your configuration

# Deploy (from agent directory)
make setup-dev-env
make deploy

# Test your deployed agent using Jupyter notebook
jupyter notebook notebooks/adk_app_testing.ipynb

Note: After successful deployment, the deployment_metadata.json file is automatically updated with your Agent Engine ID. Use the provided Jupyter notebook (notebooks/adk_app_testing.ipynb) to validate and test your deployed agent programmatically.

References

Resource	Description
Google ADK Documentation	Complete Google ADK documentation and guides
Python Quickstart	Get started with Google ADK in Python
Tools Documentation	Learn about Google ADK tools and capabilities
MCP Tools	Model Context Protocol integration guide
Google ADK Deployment	Overview of deployment options for Google ADK agents
└─ Agent Engine	Deploy agents to Vertex AI Agent Engine
└─ Cloud Run	Deploy agents to Google Cloud Run
└─ GKE	Deploy agents to Google Kubernetes Engine
Vertex AI Agent Engine	Vertex AI Agent Engine overview and documentation
Vertex AI RAG Engine	Vertex AI RAG Engine overview and documentation
Vertex AI Search	Vertex AI Search (formerly Enterprise Search) overview and documentation
Agent Starter Pack	Agent Starter Pack documentation and guides
Agent Templates Overview	Detailed information about available agent templates
Remote Templates	Guide to using Git repositories as remote templates
Agent Starter Pack GitHub	Agent Starter Pack source code and repository

Repository

This repository contains advanced implementations and examples for the Google Agent Development Kit (Google ADK).

GitHub Repository: arjunprabhulal/adk-advanced

What's Included

This workshop repository demonstrates:

Function Tools: Custom tool implementations using Google ADK
MCP Integration: Model Context Protocol (MCP) toolsets and servers
Vertex AI Agent Engine: Deployment patterns and configurations
Vertex AI RAG Engine: Retrieval-Augmented Generation implementations
Vertex AI Search: Enterprise search integration examples
Agent Starter Pack (ASP): Production-ready templates for Agent Engine, Cloud Run, and GKE deployments with built-in CI/CD, observability, and security
Demo Images: Visual demonstrations in root images/ folder showing example agent interactions

Contributing

Contributions, issues, and feature requests are welcome!

If you'd like to contribute:

Fork the repository
Create a feature branch
Commit your changes
Push to the branch
Open a Pull Request

Please ensure your code follows the existing style and includes appropriate documentation.

License

This project is part of the Google Agent Development Kit (ADK) advanced implementations.

The code examples and implementations in this repository are provided as-is for educational and demonstration purposes.

Back to Top

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
1_adk_tools		1_adk_tools
2_adk_mcp		2_adk_mcp
3_adk_deploy_agent_engine		3_adk_deploy_agent_engine
4_adk_deploy_cloudrun		4_adk_deploy_cloudrun
5_adk_deploy_gke		5_adk_deploy_gke
6_adk_deploy_rag_to_agent_engine		6_adk_deploy_rag_to_agent_engine
7_adk_vertex_ai_search_simple		7_adk_vertex_ai_search_simple
8_adk_vertex_ai_ragengine_simple		8_adk_vertex_ai_ragengine_simple
images		images
.gitignore		.gitignore
ADK-Advanced-Workshop.pdf		ADK-Advanced-Workshop.pdf
README.md		README.md

ankushkhakale/adk-advanced

Folders and files

Latest commit

History

Repository files navigation