Model Context Protocol (MCP)

Increasingly LLMs like ChatGPT have become hugely popular amongst enterprise users owing to their versatility. LLMs can be used interactively using chat interface or inside a python code or others using APis. The later is the preferred method while using AI agents.

In our case, the use of AI agents help in automation which requires integration with a LLM software. There are multiple ways in which this automation and integration can be achieved. One such method is using the MCP. The other way is using RAG.

MCP is a protocol that is implemented on a MCP server that allows LLMs to interface with external data sources like a DB or a mail application, or a help-desk application or a file. The MCP solution architecture is shown below. MCP user manual has wealth of information [1].

One can use the MCP solution with a LLM to chat with other business applications like DB/GitHub/mail/helpdesk. Claude desktop is an easy to use MCP client. Automated general purpose agentic applications can be developed. RAG solution is used to ask questions in an internal document.

A simple MCP use cause - installation

We have recreated this use case which is given in [1]. There cloud be many MCP versions. Like Open-AI, Anthropic and so on.

The MCP installation are different for each one of them. We are using Anthropic version. The following installation is  for Anthropic version.

Our environment is on Windows. We run Claude and MCP on Windows. The MCP installation On Windows was Done on GIT-BASH Terminal (LENOVO@SQLNODE1 MINGW64). using  we have  the following are MCP installation.  

Setup steps

1. Install Claude Desktop

2. Install uv by running pip install uv

3. Run uv init my-first-mcp-server to create a project directory

4. Run uv add "mcp[cli]" to add mcp cli in your project

5. Install this server inside Claude desktop by running uv run mcp run main.py & in the project directory

6. Kill any running instance of Claude from Task Manager. Restart Claude Desktop

7. In Claude desktop, now you will see tools from this server

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai

$ uv init my-first-mcp-server

Initialized project `my-first-mcp-server` at `C:\Code\course-gen-ai\my-first-mcp-server`

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai

$ cd my-first-mcp-server

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai/my-first-mcp-server (master)

$ ls

README.md main.py pyproject.toml

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai/my-first-mcp-server (master)

$ pip show mcp

Name: mcp

Version: 1.9.2

Summary: Model Context Protocol SDK

Home-page:

Author: Anthropic, PBC.

Author-email:

License: MIT

Location: C:\Python311\Lib\site-packages

Requires: anyio, httpx, httpx-sse, pydantic, pydantic-settings, python-multipart, sse-starlette, starlette, uvicorn

Required-by:

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai/my-first-mcp-server (master)

$ uv run mcp install main.py

[06/05/25 11:50:08] INFO Added server 'LeaveManager' to Claude config claude.py:143

INFO Successfully installed LeaveManager in Claude app

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai/my-first-mcp-server (master)

$ uv add "mcp[cli]"

LENOVO@SQLNODE1 MINGW64 /c/Code/course-gen-ai/my-first-mcp-server (master)

$ uv run mcp run main.py &

After installation, using Claude desktop as MCP-client, one can chat with an external data source like DB or HR system as done here.

Additional Reading

1. https://youtu.be/jLM6n4mdRuA?si=9lXHgi6P-xLlh6JE

2. https://modelcontextprotocol.io/introduction

3. Building Agents with Model Context Protocol - Full Workshop with Mahesh Murag of Anthropic, https://www.youtube.com/watch?v=kQmXtrmQ5Zg