Custommcpserver

🚀 Custom MCP Server

Build a custom MCP server using Python and connect it to an AI agent.

🚀 Quick Start

MCP server is essentially a way of communication between LLMs and AI tools. The Model Context Protocol allows applications to provide context for LLMs in a standardized way, separating the responsibility of providing context from the actual LLM interaction.

📦 Installation

# Create a project managed by uv
$ uv init .
Initialized project `custommcpserver` to `C:\Users\abhis\Desktop\AIAgents\CustomMCPServer`
# Then add MCP to the project's dependencies:
$ uv add "mcp[cli]"
# Activate the Python environment
$ source .venv/scripts/activate
# We can install this server in Claude Desktop and interact with it immediately by running the following command:
# mcp install main.py
# Since we are in VS Code, we execute
$ uv run mcp install main.py
[04/16/25 10:06:31] INFO     Added server 'Demo' to Claude config
                    INFO     Successfully installed Demo in Claude app

💻 Usage Examples

Basic Usage

• Open the Claude app > File > Settings > Developer > Edit Configuration - this will open a file, then find the claude_desktop_config.json file in it.
• We can see the path of mcp run main.py added to it.
• The MCP server should be displayed in the Claude app. If it is not displayed, do the following:
  • Close the Claude app - open the Task Manager and search for Claude, right - click and end the task.
  • Wait a few seconds and then reopen the app.
  • Once the app is loaded, it will start the MCP server and it will be displayed in the app. Refer to the images below. When you click on the tool, we can see the server: DEMO.

Advanced Usage

$ uv add black
# After creating a new MCP server and tools
$ uv run mcp install adv_main.py
[04/16/25 10:44:02] INFO     Added server 'AI Sticky Notes' to Claude config
                    INFO     Successfully installed AI Sticky Notes in Claude app

✨ Features

• Resources:
  • Python MCP Server SDK
  • Reference Code

🔧 Technical Details

• Before we proceed, what is MCP fundamentally?:

  • It enables you to build servers that provide data and functionality to LLM applications in a secure and standardized way. Imagine it's similar to a Web API, but specifically designed for LLM interactions.
    • Expose data through resources (which can be thought of as GET endpoints; they are used to load information into the LLM's context).
    • Provide functionality through tools (similar to POST endpoints; they are used to execute code or produce side - effects).
    • Define interaction patterns through prompts (reusable templates for LLM interactions).
    • And much more!

• From the above examples, we can see how to leverage the tools we created and automate many things. For example: let Claude connect to a database and create 100 fake user records in the database, instead of manually copying and pasting by ourselves.