MCP AI agent walking through

Luokun
May 14, 2025
Approx. read time: 4 mins

Introduction: The Need for MCP in AI Agents

Large Language Models (LLMs) like GPT-4, Claude, and Gemini have demonstrated remarkable reasoning and text-generation capabilities. However, they are fundamentally limited by their static training data—unable to access real-time information, execute external commands, or interact with enterprise systems. This is where the Model Context Protocol (MCP) comes in.

MCP is an open protocol designed to standardize interactions between AI models and external tools, databases, APIs, and services. Think of it as a universal USB-C port for AI, enabling seamless integration with real-world systems. Since its introduction by Anthropic in late 2024, MCP has rapidly gained adoption, with major players like Microsoft, Google, and OpenAI integrating it into their ecosystems.

In this blog, we’ll explore:

What MCP is and how it works
Key applications and real-world use cases
A hands-on example with Python code
The broader impact of MCP on AI agents

What is MCP?

MCP follows a client-server architecture, where:

MCP Host: The AI model (e.g., GPT-4, Claude) that needs external data or actions.
MCP Client: Mediates between the AI model and MCP servers.
MCP Server: Lightweight services that expose specific capabilities (e.g., database queries, API calls, file operations).

How MCP Works

The AI model sends a request (e.g., “fetch user data from PostgreSQL”).
The MCP client forwards this request to the appropriate MCP server.
The MCP server executes the task (e.g., runs a SQL query).
The response is returned to the AI model in a structured format.

This architecture allows AI agents to dynamically access live data, execute CLI commands, and interact with enterprise systems—breaking free from static knowledge limitations.

Current Applications of MCP

Azure MCP Server: AI Agents Managing Cloud Resources

Microsoft’s Azure MCP Server (now in public preview) enables AI agents to interact with Azure services like:

Azure Cosmos DB (query NoSQL databases with natural language)
Azure Storage (manage blobs, containers, and tables)
Azure Monitor (analyze logs using KQL)
Azure CLI (execute commands directly)

This allows developers to build context-aware AI agents that troubleshoot issues, deploy apps, and manage cloud resources autonomously.

GitHub Copilot + MCP: AI-Powered Development

GitHub Copilot now supports Agent Mode with MCP, meaning developers can:

Query databases directly from their IDE.
Execute shell commands via AI.
Automate CI/CD workflows.

Google’s A2A Protocol & MCP: Agent-to-Agent Communication

While MCP connects AI models to external tools, Google’s Agent-to-Agent (A2A) Protocol enables cross-platform agent collaboration. For example:

An AI agent in Salesforce can fetch data from MongoDB via A2A.
A logistics agent can coordinate with a payment agent to finalize transactions.

Enterprise AI Assistants with Real-Time Data

Companies like NineTech and BetterYeah AI use MCP to:

Fetch live financial data for trading bots.
Automate customer support with real-time CRM access.
Streamline DevOps by letting AI agents debug production logs

Hands-On Example: Building an MCP Server in Python

Let’s create a simple MCP server that performs math operations and integrates with an AI agent.

Step 1: Install Required Libraries

pip install langchain-mcp-adapters fastapi uvicorn

Step 2: Define the MCP Server

from fastapi import FastAPI
from pydantic import BaseModel
from mcp.server.fastmcp import FastMCP

app = FastAPI()
mcp = FastMCP("MathServer")

# Register a tool for addition
@mcp.tool()
def add(a: int, b: int) -> int:
    return a + b

# Register a tool for multiplication
@mcp.tool()
def multiply(a: int, b: int) -> int:
    return a * b

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)

Step 3: Connect an AI Agent to the MCP Server

from langchain.agents import AgentExecutor
from langchain.tools import Tool
from langchain.llms import OpenAI

# Initialize LLM
llm = OpenAI(model="gpt-4")

# Define MCP tool
mcp_tool = Tool(
    name="MathServer",
    func=lambda query: mcp.execute(query),
    description="Performs math operations (add, multiply)"
)

# Create an agent
agent = AgentExecutor.from_llm_and_tools(
    llm=llm,
    tools=[mcp_tool],
    verbose=True
)

# Run the agent
response = agent.run("What is (5 + 7) multiplied by 3?")
print(response)

Expected Workflow

The AI agent receives the question.
It decomposes the task into (5 + 7) = 12 and 12 * 3 = 36.
It calls the MCP server for each operation.
The final answer is returned.

The Future Impact of MCP

Standardizing AI Tool Integration

Before MCP, integrating AI with external APIs required custom code for each service. Now, MCP acts as a universal adapter, reducing development time by 73% (Gartner 2025).

Enabling Autonomous AI Agents

With MCP, AI agents can:

Self-debug code by querying logs.
Automate business workflows (e.g., invoice processing).
Control IoT devices (e.g., smart factories).

Security & Governance Challenges

MCP introduces risks like:

Unauthorized tool access (e.g., SSH keys exposed to AI).
Data leakage occurs if MCP servers aren’t properly secured. Solutions include RBAC controls and encrypted MCP channels

Key takeaways

MCP is to AI agents what REST was to web services—a simple contract that unlocks massive composability.
The ecosystem is snowballing: every major cloud and half the popular dev-tool vendors now ship first-class MCP support.
Even a “hello world” agent can list tools, write remote files and chain actions in minutes.
The upside is faster innovation; the challenge is securing and governing a protocol designed for openness.

If you’re building AI agents in 2025, learning MCP is rapidly becoming table stakes. Happy hacking!

Services

Browse our services

Learn more

Case Study

A military organisation harnesses AI