THEYASHGAUR/learning-mcp-servers

MCP Server + LLM Tool Bridge

This repo/project contains 2 phases:

Phase 1: Basic MCP Server

We built a very simple MCP style server using FastAPI. This server exposes 3 tools:

• get_weather(city)
• summarize_text(text)
• convert_currency(amount, from_currency, to_currency)

We test these tools using Postman / curl / python requests.

The server exposes a single endpoint:

POST /call_tool

where we pass:

{
  "tool": "get_weather",
  "params": { "city": "Delhi" }
}

The server receives the request → calls the correct function → returns response.

Phase 2: LLM Tool Bridge

We then created another python script which connects an LLM to the MCP server.

Flow:

• User asks a normal english question
• LLM decides which tool should be used
• We call the MCP server endpoint from python
• And return final answer to user

This is how the LLM agent + MCP style integration works.

Result

We learned how to:

• expose simple tools via a server
• test them via postman
• let an LLM act like an agent and call those tools

This is the base concept of AI Agents + MCP style architecture.

Tech used

• Python
• FastAPI
• OpenAI
• Requests
• Postman

Run server

uvicorn main:app --reload

Run LLM client

python llm_client.py

This is a great base for building more complex MCP tool networks later.

MCP Server Project - Phase 1 & Phase 2 README

Overview

This project demonstrates building a simple local MCP server that exposes basic capabilities, and then expands it by connecting it to an external API. It is divided into 2 Phases to help you understand step–by–step.

Phase 1 (Local basic MCP Server)

What happens in Phase 1?

• You create a basic MCP server using the mcp python package.
• You expose 2 simple endpoints that return static JSON responses.
• These endpoints behave like mini tools.
• There is no external API involved in Phase 1.

Goal of Phase 1: Understand how MCP server works, how tools are exposed, and how the MCP protocol responds.

Phase 2 (Connecting external API - Joke API)

What happens in Phase 2?

• You extend the same MCP server logic.
• You call a third‑party public Joke API.
• The MCP server returns a real API response instead of static JSON.

Goal of Phase 2: Learn how to call external APIs using fetch (python requests) and send dynamic results back through MCP.

Tech used

• Python
• mcp python lib
• FastAPI (optional) for checking via POSTMAN

How to test using Postman

• You run your python server using: python server.py

• The server exposes an HTTP endpoint (FastAPI) for testing

• You can do a POST request on Postman:

  {
    "tool": "getJoke"
  }
  

• You get JSON response back from MCP server

Summary

This project shows clear evolution from MCP basics → to real use case integration.

Next Phase ideas: integrate DB, RabbitMQ, Redis, Billing, Analytics