CustomMCPServer

CustomMCPServer

Build a custom MCP server in Python and connect that to an AI agent.

• Resources:

  • Python MCP Server SDK
  • Reference Code

• MCP servers are essentially how LLM and AI Tools communicate with each other.The Model Context Protocol allows applications to provide context for LLMs in a standardized way, separating the concerns of providing context from the actual LLM interaction.

# create a uv-managed project
$ uv init .
Initialized project `custommcpserver` at `C:\Users\abhis\Desktop\AIAgents\CustomMCPServer`
# Then add MCP to your project dependencies:
$ uv add "mcp[cli]"
# activate python environment as well
$ source .venv/scripts/activate
# You can install this server in Claude Desktop and interact with it right away by running:
# mcp install main.py
# since we're in this vs code we do
$ 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
# go to claude app > File > settings > Developer > Edit Config - opens up some file and open that claude_desktop_config.json file
# we can see the mcp run main.py(path) is added in that file.
# the mcp server should show up on claude app - if it's not - we do the following
# close claude app - open task manager and search fo claude and right click & do end task.
# wait a sec & open the app again
#  once the app loads , it spins up the mcp server and it shows on app - refer the image below & when you click on tool we can see the
# add tool and server:DEMO

Advanced MCP Server

• before we go ahead , fundamentally What is MCP?:
  • lets you build servers that expose data and functionality to LLM applications in a secure, standardized way. Think of it like a web API, but specifically designed for LLM interactions.
    • Expose data through Resources (think of these sort of like GET endpoints; they are used to load information into the LLM's context)
    • Provide functionality through Tools (sort of like POST endpoints; they are used to execute code or otherwise produce a side effect)
    • Define interaction patterns through Prompts (reusable templates for LLM interactions)
    • And more!

uv add black
# after creating new mcp server & tool
$ 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

• from this above example , we can see how to make use of tool we create and do whatever we can like.
• add a tool to connect to db and ask claude to create 100 fake users into db , instead of us manually copy paste ..etc.we can automate lot of stuff.