Build Your Own MCP Server in Python & Connect It to Gemini AI

Large Language Models (LLMs) are exceptionally powerful at reasoning and text generation, but they are naturally isolated from your local data, enterprise tools, and custom backend APIs.

The Model Context Protocol (MCP) bridges this gap by providing an open, standardized architecture that lets AI models interact securely with external context, tools, and databases without relying on brittle, custom integration scripts.

In this guide, you will learn what MCP is, why it was implemented, how to build a local MCP server in Python, and how to connect it directly to Google’s Gemini API.

Also for more interesting blogs on Artificial Intelligence have a look at

https://amplifyabhi.com/category/artificial-intelligence

also youtube playlist Artificial Intelligence Guide

What is MCP & Why Was It Implemented?

The Core Problem: The N * M Integration DilemmaBefore MCP, every AI provider (Anthropic, OpenAI, Google) used proprietary API formats to interface with external tools (such as GitHub, SQL databases, or Figma).

If you had N AI models and M tools, developers had to build and maintain N * M custom integration pipelines.These legacy integrations were brittle, non-standardized, and created severe information silos.

To have the best understanding we also created a MCP Server in Python step by step to let you know and understand behind the scenes what exactly happens.

The Solution: A Unified Protocol

Introduced by Anthropic and maintained under the Linux Foundation’s Agentic AI Foundation, MCP standardizes how AI hosts communicate with context servers.

• Standardized Capabilities: Exposes tools, resources, and prompt templates to AI models through a uniform client-server protocol.
• Security & Isolation: Executes tasks securely via defined transports (such as stdio or HTTP/SSE), keeping execution logic isolated from model context.
• Reusability: Build an MCP tool once, and it instantly works across any compliant client—such as Claude Desktop, Cursor, or custom Python clients running Gemini.

Popular MCP Tools in the Ecosystem

While you can build custom servers, a vast open-source ecosystem of pre-built MCP servers already exists:

Popular MCP Server	Core Capability	Primary Use Case
GitHub / Git MCP	Repository management, file reading, PR inspection	Automated code reviews, git operations, PR summaries
PostgreSQL / Supabase	Read-only database queries, schema inspection	Instant SQL querying and dynamic context retrieval
Filesystem MCP	Secure, scoped local file read/write operations	Automated file modifications and folder inspection
Figma MCP	Extracts components, design tokens, and assets	UI/UX design-to-code generation workflows
Slack / Sentry MCP	Reads channel logs and issue error traces	Live system diagnostic monitoring and AI ops

Project Requirements & Setup

Prerequisites

• Python 3.10 or higher installed.
• A Gemini API key (obtainable for free via Google AI Studio).

1. File Structure

Organize your workspace into three primary files:

mcp-gemini-demo/
├── .env
├── requirements.txt
├── main.py        # Direct Gemini API integration test
├── server.py      # Custom Local MCP Server definition
└── client.py      # Client connecting Gemini to the local MCP Server

2. Dependencies (requirements.txt)

These are the required dependecies to create MCP Server in Python.

google-genai
python-dotenv
mcp

Install dependencies via your terminal:

pip install -r requirements.txt

3. Environment Variables (.env)

Configure your secret keys in this file explained in detail in video tutorial.

GEMINI_API_KEY=your_actual_gemini_api_key_here

4. How-To Guide: Step-by-Step Implementation

Step 1: Testing Basic Gemini Integration (main.py)

This script verifies that your GEMINI_API_KEY is loaded correctly and that the SDK can communicate with Google LLM models.

# Quick snippet: Generating content via Gemini API
client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))

response = client.models.generate_content(
    model="gemini-3.6-flash",
    contents="Explain Android Clean Architecture"
)
print(response.text)

Step 2: Custom Local MCP Server (server.py)

Expose functions using the @mcp.tool() decorator to mark them for model recognition.

# Quick snippet: Decorating custom tools for MCP discovery
mcp = MCPServer("My Local Server")

@mcp.tool()
def add(a: int, b: int) -> int:
    """Adds two numbers (or modified local execution logic)."""
    return a * b  # Tested with multiplication to confirm local tool execution

if __name__ == "__main__":
    mcp.run(transport="stdio")

Step 3: MCP Client Session Execution (client.py)

Launch the server sub-process, list tools, and execute a tool via ClientSession.

# Quick snippet: Launching stdio connection and invoking a tool
async with stdio_client(server_params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        
        # Call tool directly from client session
        result = await session.call_tool("add", {"a": 3, "b": 4})
        print("Tool result:", result)

5. Troubleshooting Guide

Issue / Error	Cause	Recommended Solution
FileNotFoundError: server.py	client.py cannot locate the server script relative to the execution path.	Pass absolute paths in StdioServerParameters or verify your current working directory.
TypeError: dict expected for arguments	Parameters passed to session.call_tool() are not formatted as a JSON dictionary.	Ensure arguments match key-value mapping (e.g., {"a": 3, "b": 4}).
API Key Invalid / Not Found	Environment variable GEMINI_API_KEY failed to load before client initialization.	Ensure load_dotenv() is called at the top of the file before genai.Client() runs.
Stdio Transport Freeze	Server script contains print() statements that output raw text into the stdio pipe.	Remove arbitrary print() calls from server.py to prevent breaking stdio protocol frames.

MCP Server in Python :

Go through the below tutorial to get complete understanding of creating your own MCP Server in Python.

Complete code for creating MCP Server in Python will be uploaded into github.

Let us know your feedback and queries in below comment section.