MCP Projects

🚀 mcp-projects

This is the repository for my MCP (Model Context Protocol) project, which aims to provide a standardized way for applications to offer richer context to AI models.

🚀 Quick Start

📦 Installation

1. Clone this repository.
2. Install the dependencies:

pip install mcp
pip install fastapi
pip install uvicorn
pip install fastapi-mcp
pip install llama-index
pip install llama-index-embeddings-huggingface
pip install llama-index-llms-langchain
pip install langchain-mcp-adapters
pip install mcp-use

3. Create a .env file in the root folder with the following credentials:

API_KEY=<IBM Cloud API Key>
PROJECT_ID=<Watsonx Project ID>
IBM_CLOUD_URL=<IBM Cloud URL>

MODEL_ID=<Your Watsonx.ai LLM ID>

TAVILY_API_KEY=<Your Tavily Web Search API Key>

Alternatively, use your own LLM provider - it's agnostic to the project (some changes will be required).

4. Experiment with different projects and files:
   • Make sure to run the MCP server first, then the client.
   • Run the client.

✨ Features

What is MCP?

Fundamentally, MCP is a standardized method that enables applications to provide their environment, user preferences, and conversation history to AI models in a more comprehensive way. Think of it as a means to offer intelligent, structured context feeds to AI systems.

Problems Solved by MCP

Current AI systems have limited "working memory" - they can only view a specific amount of conversation history at a time (their "context window"). Imagine having a conversation with someone who can only remember the most recent exchanges:

• You: "Remember that project we discussed last week about optimizing the supply chain?"
• AI without good context: "I don't recall that specific discussion. Can you remind me of the details?"

This limitation forces users to repeatedly explain content, resulting in a frustrating interaction experience. MCP addresses this issue by creating a structured method for maintaining and accessing context.

Analogy

1. GPS Navigation

Traditional AI context management is like someone who only gives one turn direction at a time without showing the full map. If they forget a step, the entire conversation breaks down.

MCP is like a GPS navigation system:

• It knows your destination.
• It remembers your preferred routes.
• It adjusts according to real - time conditions.
• It always knows your position in the journey.

Read this article for a comprehensive understanding of MCP: Understanding Model Context Protocol (MCP): A Layman's Guide

💡 Usage Tip

Please contribute to this asset via Pull Request ✨