AADK - AI Agent Development Kit Demo

A comprehensive showcase of AI agents built with Google ADK (Agent Development Kit), demonstrating various agent architectures and integrations including MCP tools, A2A protocol, BigQuery, and multi-agent workflows.

Overview

This project demonstrates 5 different AI agent implementations, each showcasing different capabilities of the Google ADK framework:

• MCP Agent - Multi-tool integration using Model Context Protocol
• ArXiv Agent - Sequential multi-agent workflow for research paper analysis
• BigQuery Agent - Database querying with vision capabilities
• A2A Agent - Agent-to-Agent communication across different languages
• Sustainability Research Agent - Parallel research execution and synthesis

Agents

🔧 MCP Agent

An Agent Navigator that integrates with multiple open-source tools via Model Context Protocol (MCP).

Connected Tools:

• GitHub Copilot MCP
• Hugging Face MCP
• Context7 MCP

Capabilities: Repository search, model exploration, and contextual information retrieval with multilingual support.

📄 ArXiv Agent

A sequential agent workflow that searches and reviews academic papers from arXiv.

Architecture:

• scraper_agent - Fetches latest papers (max 7) based on domain
• reviewer_agent - Analyzes and reviews paper content

Output: Structured paper summaries with PDF links and analysis.

💾 BigQuery Agent

A dual-mode intelligent assistant combining data analytics with device assessment.

Modes:

1. Senior Data Architect - Writes optimized GoogleSQL queries
2. Device Assessment Specialist - Grades and values physical devices (iPhone trade-ins)

Features:

• BigQuery integration with write protection
• Custom phone grading tool with vision capabilities
• Context-aware routing logic

🎲 A2A Agent

Demonstrates Agent-to-Agent (A2A) protocol communication between Python and Go agents.

Components:

• Python agent orchestrator
• Remote Go agent (check_prime_agent) for prime number verification
• Dice rolling sub-agent

Use Case: Cross-language agent collaboration and remote agent integration.

🌱 Sustainability Research Agent

Parallel research agent that investigates multiple sustainability topics simultaneously.

Research Topics:

• Renewable Energy
• Electric Vehicle Technology
• Carbon Capture Methods

Architecture: Uses ParallelAgent to run research tasks concurrently, then merges results.

Project Structure

aadk/
├── agents/
│   ├── a2a_agent/           # Agent-to-Agent communication
│   │   └── remote_a2a/
│   │       └── check_prime_agent/  # Go-based remote agent
│   ├── arxiv_agent/         # Sequential paper research
│   │   └── sub_agents/
│   ├── bq_agent/            # BigQuery database agent
│   │   ├── prompts/
│   │   └── tools/
│   ├── mcp_agent/           # MCP tool integrations
│   └── sustainability_research_agent/  # Parallel research
├── shared/
│   └── config.py           # Shared configuration
├── pyproject.toml
├── Makefile
└── README.md

Requirements

• Python >= 3.12
• Go (for A2A remote agent)
• Google Cloud Project with BigQuery access

Installation

Option 1: Using uv (Recommended)

uv is a fast Python package installer and resolver.

1. Install uv

# On Linux/Mac
curl -LsSf https://astral.sh/uv/install.sh | sh

# Or using pip
pip install uv

2. Clone and setup

git clone <repository-url>
cd aadk

3. Create virtual environment and install dependencies

uv venv
source .venv/bin/activate  # On Linux/Mac
uv pip install -e .

Option 2: Using pip

1. Clone the repository

git clone <repository-url>
cd aadk

2. Create virtual environment

python -m venv .venv
source .venv/bin/activate  # On Linux/Mac

3. Install dependencies

pip install -e .

Configuration

4. Configure environment variables Create a .env file with required credentials:

# API Tokens
GITHUB_TOKEN=your_github_token
HUGGING_FACE_TOKEN=your_hf_token
NOTION_TOKEN=your_notion_token
CONTEXT7_TOKEN=your_context7_token

# Google Cloud
GOOGLE_GENAI_USE_VERTEXAI=FALSE
GOOGLE_CLOUD_PROJECT=your_project_id
GOOGLE_CLOUD_LOCATION=your_location

# BigQuery
BQ_DATASET_NAME=your_dataset
BQ_DATASET_LOCATION=your_location
BQ_TABLE=your_table
BQ_TUKERINAJA_TABLE=your_table

# GCS (if needed)
GCS_BUCKET_NAME=your_bucket
GCS_BLOB_NAME=your_blob

# Model
MODEL_ID=gemini-2.0-flash-001

Usage

Start All Agents

make start

This will:

1. Start the Go-based remote A2A agent
2. Launch the ADK web interface for all agents

Run Individual Agents

Using the ADK CLI:

# Run specific agent
adk agent run agents/mcp_agent

# Run with web interface
adk web agents/mcp_agent

Dependencies

Core:

• google-adk[a2a] - Google Agent Development Kit
• a2a-sdk[all] - Agent-to-Agent SDK
• langchain - LLM framework
• langchain-google-genai - Google Gemini integration

Data & Processing:

• pandas - Data manipulation
• google-cloud-bigquery - BigQuery client
• db-dtypes - Database type handling

Utilities:

• bs4 - Web scraping
• pypdf - PDF processing

Key Features

✅ Multi-Agent Architectures - Sequential, Parallel, and Remote agent patterns
✅ MCP Integration - Model Context Protocol for tool interoperability
✅ Cross-Language A2A - Python-Go agent communication
✅ Vision Capabilities - Multimodal phone grading
✅ BigQuery Integration - Secure database querying
✅ Research Automation - Parallel web research and synthesis

License

Apache License 2.0 (see agent files for full license text)

Architecture Patterns

Agent	Pattern	Communication
MCP Agent	Single Agent	HTTP/MCP
ArXiv Agent	Sequential	Internal
BQ Agent	Single Agent	BigQuery API
A2A Agent	Remote/Hybrid	A2A Protocol
Sustainability Agent	Parallel + Sequential	Internal

Development

Adding a new agent:

1. Create agent directory under agents/
2. Implement agent.py with root_agent export
3. Add __init__.py with settings import
4. Configure in shared settings if needed

Agent Template:

from google.adk.agents import Agent

root_agent = Agent(
    name="my_agent",
    model="gemini-2.0-flash-001",
    instruction="Your system instruction",
    tools=[],
)

---

Built with Google ADK 🚀