15-mcp-deep-dive

15. Model Context Protocol (MCP) Deep Dive

MCP architecture, connection types, and production patterns.

Blog Post: https://arjunprabhulal.com/adk-mcp-deep-dive/

Table of Contents

1. What is MCP?
2. Prerequisites
3. Setup Steps
4. Connection Types
5. Running the Agent
6. Next Steps

What is MCP?

Model Context Protocol is a standardized way for AI agents to connect to external services and data sources. Key benefits:

• Standardized interface - One protocol for many services
• Dynamic tool discovery - Agents learn available tools at runtime
• Secure connections - Built-in authentication support

Prerequisites

• Python 3.10+
• Gemini API key from AI Studio
• GitHub Personal Access Token
• Node.js (for Firecrawl MCP - optional)

Setup Steps

1. Navigate to this module:

cd 15-mcp-deep-dive

2. Create and activate a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

pip install -r ../requirements.txt

4. Set up environment variables in mcp_agent/.env:

GOOGLE_API_KEY=your-api-key-here
GITHUB_TOKEN=your-github-token
FIRECRAWL_API_KEY=your-firecrawl-api-key  # Optional

Connection Types

Type	Class	Use Case
HTTP	StreamableHTTPServerParams	Cloud-hosted MCP servers (GitHub Copilot MCP)
Stdio	StdioConnectionParams	Local process-based servers (npx tools)

HTTP Connection Example

StreamableHTTPServerParams(
    url="https://api.githubcopilot.com/mcp/",
    headers={"Authorization": f"Bearer {token}"},
)

Stdio Connection Example

StdioConnectionParams(
    server_params=StdioServerParameters(
        command="npx",
        args=["-y", "firecrawl-mcp"],
    ),
)

Running the Agent

Using Programmatic Runner

cd mcp_agent
python agent.py

Test Queries:

• "Explain MCP and its use in ADK"
• "Find ADK repositories on GitHub"

Next Steps

Continue to 16. Session, State & Memory