13-third-party-mcp-tools

13. Third-Party MCP Tools

GitHub and Firecrawl integration via Model Context Protocol.

Blog Post: https://arjunprabhulal.com/adk-third-party-tools-github/

Table of Contents

1. Overview
2. Prerequisites
3. Setup Steps
4. GitHub MCP Setup
5. Running the Agent
6. Next Steps

Overview

Model Context Protocol (MCP) enables agents to connect to external services through a standardized interface. This module demonstrates GitHub integration.

Prerequisites

• Python 3.10+
• Gemini API key from AI Studio
• GitHub Personal Access Token

Setup Steps

1. Navigate to this module:

cd 13-third-party-mcp-tools

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 github_agent/.env:

GOOGLE_API_KEY=your-api-key-here
GITHUB_TOKEN=your-github-token

GitHub MCP Setup

1. Create a GitHub token:

   • Go to github.com/settings/tokens
   • Generate a new token with repo scope

2. Configure MCPToolset:

from google.adk.tools.mcp_tool.mcp_toolset import MCPToolset
from google.adk.tools.mcp_tool.mcp_session_manager import StreamableHTTPServerParams

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

Running the Agent

Using Programmatic Runner

cd github_agent
python agent.py

Using ADK Web

adk web

Test Queries:

• "List trending Python repositories"
• "Get the README from google/adk-python"

Next Steps

Continue to 14. MCP Toolbox for Databases