What is Mcp Agent Google Adk

mcp-agent-google-adk is a powerful MCP (Model Context Protocol) Agent built using Google’s Agent Development Kit (ADK) that integrates multiple data sources and services through MCP toolsets.

Use cases

Use cases include building intelligent agents for querying PostgreSQL databases, providing real-time weather updates, and connecting to remote services for enhanced functionality.

How to use

To use mcp-agent-google-adk, clone the repository, set up a Python virtual environment, install dependencies, configure environment variables for Google API, PostgreSQL, and remote services, and then run the agent.

Key features

Key features include multi-modal capabilities powered by Gemini 2.0 Flash, direct PostgreSQL database integration, real-time weather data access, remote service integration via SSE, an extensible architecture for adding new toolsets, and secure configuration management.

Where to use

mcp-agent-google-adk can be used in various fields such as data analysis, real-time weather monitoring, and integration of remote services for intelligent applications.

MCP Agent Google ADK

A powerful MCP (Model Context Protocol) Agent built using Google’s Agent Development Kit (ADK) that integrates multiple data sources and services through MCP toolsets.

Overview

This project demonstrates how to build an intelligent agent using Google ADK that can interact with:

• PostgreSQL databases via MCP server
• Weather services for real-time weather data
• Remote services through SSE connections

The agent uses Google’s Gemini 2.0 Flash model to provide intelligent responses and can execute complex queries across multiple data sources.

Features

• 🤖 Multi-Modal Agent: Powered by Gemini 2.0 Flash for advanced reasoning
• 🗄️ Database Integration: Direct PostgreSQL access through MCP
• 🌤️ Weather Data: Real-time weather information
• 🔗 Remote Service Integration: SSE-based remote service connections
• 🛠️ Extensible Architecture: Easy to add new MCP toolsets
• 🔐 Secure Configuration: Environment-based secret management

Prerequisites

• Python 3.8+
• Node.js and npm (for MCP servers)
• PostgreSQL database (if using database features)
• Google API credentials
• API access (for remote services)

Installation

1. Clone the Repository

git clone https://github.com/yourusername/mcp-agent-google-adk.git
cd mcp-agent-google-adk

2. Set Up Python Environment

# Create a virtual environment
python -m venv venv

# Activate virtual environment
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

# Install dependencies
pip install -r requirements

3. Environment Configuration

Create a .env file in the root directory:

# Google API Configuration
GOOGLE_API_KEY=your_google_api_key_here

# PostgreSQL Configuration
POSTGRES_CONNECTION_STRING=username:password@localhost:5432/database_name

# Remote Service Configuration
CONNECTION_URL=your_sse_connection_url
API_KEY=your_api_key

4. Install Node.js Dependencies

The MCP servers require Node.js packages that are installed automatically when the agent starts:

• @modelcontextprotocol/server-postgres - PostgreSQL MCP server
• @h1deya/mcp-server-weather - Weather MCP server

Usage

Basic Usage

from adk_agent_samples.mcp_agent import agent

# The agent is automatically configured and ready to use
response = agent.root_agent.run("What's the weather like today?")
print(response)

Database Queries

# Query your PostgreSQL database
response = agent.root_agent.run("Show me the latest records from the users table")
print(response)

Multi-Source Queries

# Combine data from multiple sources
response = agent.root_agent.run(
    "Get the current weather and also check if there are any users in our database from that city"
)
print(response)

Architecture

The agent is built with three main MCP toolsets:

1. PostgreSQL Toolset (mcp_pg)

• Connection: Stdio-based MCP server
• Purpose: Direct database access and querying
• Server: @modelcontextprotocol/server-postgres

2. Weather Toolset (mcp_weather)

• Connection: Stdio-based MCP server
• Purpose: Real-time weather data retrieval
• Server: @h1deya/mcp-server-weather

3. Remote Toolset (mcp_remote)

• Connection: SSE (Server-Sent Events) based
• Purpose: Integration with external services
• Authentication: API key and account ID headers

Configuration

Environment Variables

Variable	Description	Required
GOOGLE_API_KEY	Google API key for Gemini model	Yes
POSTGRES_CONNECTION_STRING	PostgreSQL database connection	Yes
CONNECTION_URL	SSE server URL for remote services	Yes
API_KEY	Service API key	Yes

Agent Configuration

The agent is configured with:

• Model: gemini-2.0-flash
• Name: mcp_agent
• Instruction: “You are an assistant for the given data.”
• Tools: All three MCP toolsets

Development

Project Structure

mcp-agent-google-adk/
├── adk_agent_samples/
│   └── mcp_agent/
│       ├── __init__.py
│       └── agent.py          # Main agent configuration
├── .env                      # Environment variables (create this)
├── .gitignore               # Python gitignore
├── requirements             # Python dependencies
└── README.md               # This file

Adding New MCP Toolsets

To add a new MCP toolset:

1. Create the toolset:

new_toolset = MCPToolset(
    connection_params=StdioServerParameters(
        command="npx",
        args=["-y", "@your/mcp-server-package"]
    )
)

2. Add to agent tools:

root_agent = LlmAgent(
    name="mcp_agent",
    model="gemini-2.0-flash",
    tools=[mcp_pg, mcp_weather, mcp_remote, new_toolset],  # Add here
    instruction="You are an assistant for the given data.",
)

Environment Setup for Development

# Install development dependencies
pip install -e .

# Set up pre-commit hooks (recommended)
pip install pre-commit
pre-commit install

Troubleshooting

Common Issues

Issue: Google API Key not working

Error: Authentication failed
Solution: Ensure your GOOGLE_API_KEY is valid and has access to Gemini API

Issue: PostgreSQL connection failed

Error: Connection refused
Solution: Check your POSTGRES_CONNECTION_STRING format and database accessibility

Issue: MCP server installation fails

Error: npx command not found
Solution: Install Node.js and npm, ensure they're in your PATH

Issue: Remote service connection timeout

Error: SSE connection failed
Solution: Verify CONNECTION_URL and API_KEY are correct

Debug Mode

To enable verbose logging, set environment variable:

export GOOGLE_ADK_DEBUG=true

Dependencies

Core Dependencies

• google-adk==1.0.0 - Google Agent Development Kit
• python-dotenv==1.1.0 - Environment variable management
• fastapi==0.115.12 - Web framework (if needed)
• httpx==0.28.1 - HTTP client for API calls

Google Cloud Dependencies

• google-genai==1.16.1 - Google Generative AI
• google-cloud-aiplatform==1.94.0 - AI Platform integration
• Various other Google Cloud services

MCP Dependencies

• mcp==1.9.1 - Model Context Protocol implementation
• httpx-sse==0.4.0 - Server-Sent Events support

Resources

• Google ADK Documentation
• Model Context Protocol
• Gemini API Documentation
• MCP Server Registry

Support

For questions and support:

• Create an issue on GitHub
• Check the Google ADK documentation
• Review the MCP server documentation for specific toolset issues

---

Built with ❤️ using Google ADK and Model Context Protocol (MCP)