mcp-weather-server

MCP Weather Application (Client-Server)

This project demonstrates a client-server application using the MCP (Model Context Protocol) framework to retrieve weather information.

• weather/: Contains the MCP server code, which fetches data from the National Weather Service (NWS) API.
• mcp-client/: Contains the MCP client code, which interacts with the server and uses LangChain with Groq for natural language queries.

Description

The MCP Server (weather/) provides specific tools (get_alerts, get_forecast) accessible via MCP. It connects to the NWS API to fulfill requests made through these tools.

The MCP Client (mcp-client/) connects to the MCP server, discovers its tools, and allows users to interact via a chat interface. It uses LangChain and the Groq API (Llama 3 model) to process natural language queries, determine if a server tool is needed, call the tool via MCP if necessary, and formulate a final response.

Architecture

1. The Client starts and initiates an MCP connection (via stdio) to the Server script.
2. The Client discovers the tools available on the Server.
3. The user provides a natural language query to the Client.
4. The Client uses the Groq LLM (via LangChain) to interpret the query and potentially identify the need to use a server tool.
5. If a tool is needed, the Client sends an MCP request to the Server to execute the specific tool (e.g., get_forecast) with the required arguments.
6. The Server executes the tool logic, fetching data from the NWS API.
7. The Server sends the result back to the Client via MCP.
8. The Client incorporates the tool's result into the ongoing conversation with the LLM.
9. The Client presents the final LLM-generated response to the user.

Setup

Detailed setup instructions, including dependency installation and environment configuration (like API keys), can be found in the README files within each component's directory:

• Server Setup: See weather/README.md
• Client Setup: See mcp-client/README.md

Running the Application

To run the client and have it automatically start and connect to the server, execute the following command from the mcp-client/ (the directory containing client.py):

uv run client.py ../weather/weather.py