Model Context Protocol (MCP)

Installing Dependencies

First, install the dependencies:

uv install -r requirements.txt

Running the Server

To run the server, execute the following command:

uv run main.py

After running the server, you must to use or MCP Inspector or any other client to interact with the server. The server will be listening on http://localhost:8000/mcp.

To run the MCP Inspector, execute the following command:

npx -y @modelcontextprotocol/inspector

Set the URL to http://localhost:8000/mcp and click "Connect". You should see the server's capabilities and be able to send requests to it.

Concepts

Inside the MCP, we have three main concepts: Resources, Tools, and Prompts.

• Resources are used to represent data that can be accessed by the model. They are defined using the @mcp.resource decorator and can be accessed using a URL-like syntax. For example, greetings://{name} is a resource that generates a personalized greeting based on the provided name.
• Tools are functions that perform specific actions or calculations. They are defined using the @mcp.tool decorator and can be called by the model to perform tasks. For example, the add tool takes two numbers and returns their sum.
• Prompts are used to generate prompts for the model. They are defined using the @mcp.prompt decorator and can be used to create dynamic prompts based on input parameters. For example, the calculate_prompt prompt generates a prompt for performing a mathematical operation based on the provided operation and numbers.

Usually after call a prompt, it will redirect to a resource or a tool, but it can also return a string that can be used as a prompt for the model.