🚀 Building-ADK-Agents-with-MCP-Integration

📌 Overview

This project implements Model Context Protocol (MCP) with Google Agent Development Kit (ADK) to:

• Use external MCP tools (Google Maps)
• Build a custom MCP server
• Enable tool-based AI agent workflows

This project is based on the official Google Skills lab GENAI124.

🧠 What is MCP?

Model Context Protocol (MCP) is an open standard that enables LLMs to interact with external tools, APIs, and data sources using a client-server architecture.

🎯 Objectives

• Use ADK agent as MCP client
• Integrate Google Maps MCP server
• Build custom MCP server
• Enable tool execution via agents

📂 Project Structure

adk_mcp_tools/
│
├── adk_mcp_server/
├── google_maps_mcp_agent/
├── callback_logging.py
├── requirements.txt

adk_utils/
├── plugins.py

⚙️ Setup & Installation (Task 1 - Lab)

Step 1: Download Lab Files

gcloud storage cp -r gs://YOUR_GCP_PROJECT_ID-bucket/* .

Step 2: Install ADK

export PATH=$PATH:"/home/${USER}/.local/bin"
python3 -m pip install google-adk -r adk_mcp_tools/requirements.txt

🔑 Task 2: Google Maps MCP Integration

Step 1: Create API Key

• Go to Google Cloud Console → Credentials
• Click Create API Key
• Rename it to:

GOOGLE_MAPS_API_KEY

Step 2: Create .env File

cd ~/adk_mcp_tools
cat << EOF > google_maps_mcp_agent/.env
GOOGLE_GENAI_USE_VERTEXAI=TRUE
GOOGLE_CLOUD_PROJECT=Project
GOOGLE_CLOUD_LOCATION=global
GOOGLE_MAPS_API_KEY="YOUR_API_KEY"
MODEL=gemini_flash_model_id
EOF

Step 3: Copy .env

cp google_maps_mcp_agent/.env adk_mcp_server/.env

Step 4: Add MCPToolset in agent.py

tools=[
    MCPToolset(
        connection_params=StdioConnectionParams(
            server_params=StdioServerParameters(
                command='npx',
                args=[
                    "-y",
                    "@modelcontextprotocol/server-google-maps",
                ],
                env={
                    "GOOGLE_MAPS_API_KEY": google_maps_api_key
                }
            ),
            timeout=15,
        ),
    )
]

Step 5: Run ADK UI

cd ~/adk_mcp_tools
adk web --allow_origins "regex:https://.*\.cloudshell\.dev"

Step 6: Open UI

http://127.0.0.1:8000

Step 7: Test Prompts

Get directions from GooglePlex to SFO

What's the route from Paris, France to Berlin, Germany?

🧩 Task 3: Custom MCP Server

Step 1: Update Path in agent.py

PATH_TO_YOUR_MCP_SERVER_SCRIPT = "/home/USERNAME/adk_mcp_tools/adk_mcp_server/adk_server.py"

Step 2: Add MCPToolset

tools=[
    MCPToolset(
        connection_params=StdioConnectionParams(
            server_params=StdioServerParameters(
                command="python3",
                args=[PATH_TO_YOUR_MCP_SERVER_SCRIPT],
            ),
            timeout=15,
        ),
        tool_filter=['load_web_page']
    )
]

Step 3: Run MCP Server

python3 ~/adk_mcp_tools/adk_mcp_server/adk_server.py

Step 4: Start ADK UI (New Terminal)

cd ~/adk_mcp_tools
adk web --allow_origins "regex:https://.*\.cloudshell\.dev"

Step 5: Test Agent

Select: adk_mcp_server

Load the content from http://example.com

🌍 Running on Personal Google Cloud Account (Extra Setup)

🔧 What You Need To Set Up

1. Enable Billing 💰

• Go to Billing → Enable billing
• Required for API usage

2. Enable Required APIs

Go to APIs & Services → Library and enable:

• Vertex AI API
• Directions API
• Routes API
• (Optional) Maps JavaScript API

3. Create API Key

• Go to Credentials → Create API Key

• Restrict it to:

  • Maps APIs (Directions / Routes)

4. Set IAM Permissions

Ensure your account has:

• Owner / Editor role
• OR required service roles

5. Install ADK Locally

pip install google-adk

6. Update .env

GOOGLE_GENAI_USE_VERTEXAI=TRUE
GOOGLE_CLOUD_PROJECT=your_project_id
GOOGLE_CLOUD_LOCATION=global
GOOGLE_MAPS_API_KEY=your_api_key
MODEL=gemini-1.5-flash

7. Run Project

cd adk_mcp_tools
adk web

🔄 How It Works

Flow:

1. ADK Agent → MCPToolset
2. MCPToolset → MCP Server
3. Tool execution (load_web_page / Maps)
4. Response → Agent

✨ Key Features

• Google Maps integration
• Custom MCP server
• Tool-based AI interaction
• Standardized MCP communication

🏁 Conclusion

This project demonstrates how MCP enables scalable AI systems by connecting agents with external tools and services using ADK.