mcp_demos

rpurser47/mcp_demos

3.2

If you are the rightful owner of mcp_demos and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

The Weather Chatbot is a Streamlit-based application that uses a local Llama 3.2 model and an MCP server to provide live weather data.

Tools
1
Resources
0
Prompts
0

Weather Chatbot (Llama 3.2 + MCP)

A robust, maintainable Streamlit-based weather chatbot powered by local Llama 3.2 via Ollama and live weather data from your MCP server (National Weather Service API).

Features

  • Natural chat interface with Streamlit
  • Local Llama 3.2 model (Ollama)
  • Tool-calling for live weather (via MCP server)
  • Modular, easy-to-extend codebase (LangGraph)

Setup

  1. Install Python 3.13+
  2. Clone this repo
  3. Install dependencies: 
    pip install -r requirements.txt
  4. Start your MCP server (see mcp_nws directory)
  5. Start Ollama with Llama 3.2 model: 
    ollama run llama3.2
  6. Run the app: 
    streamlit run streamlit_app.py

Configuration

  • Edit chatbot_system_prompt.md to change the chatbot's behavior.
  • All weather queries require the MCP server running on localhost:8000.

Usage

  • Ask about the weather in any US location.
  • The bot will call the weather tool for live data and answer based on the result.

What You'll Find Here

  • MCP Server: A FastAPI-based server that provides structured access to current and forecast weather data from the NWS, with endpoints designed for both LLMs and traditional applications.
  • Demo Application: (Optional/future) A conversational interface where users can ask about the weather in natural language, with the LLM leveraging the MCP server for live data.
  • Learning Focus: This project is intended as a practical introduction to MCP concepts, showing how external APIs can be made accessible to LLMs and other intelligent agents.

Project Structure

  • mcp_nws/ — Contains all code, the detailed API README, and implementation for the MCP server.
  • PLAN.md — The project plan and design rationale.

Getting Started

1. Download and Install Ollama (for Llama 2)

  • Visit https://ollama.com/download and follow the instructions for your operating system.
  • After installation, open a terminal and run: 
    ollama pull llama2
    This will download the Llama 2 model for local use.

2. Start the Ollama Server

  • In a terminal, run: 
    ollama serve
    This will start the Ollama server on localhost:11434 (the default port).

3. Start the MCP Server

  • In your project root, start the MCP server (FastAPI/Uvicorn) with: 
    uvicorn mcp_nws.main:app --reload
    This will start the MCP server on localhost:8000.

4. Start the Demo App (Streamlit)

  • In your project root, run: 
    streamlit run streamlit_app.py
    The app will be available at http://localhost:8501.

For setup instructions, API endpoint documentation, and usage examples, see:

👉

This file contains everything you need to run, test, and extend the MCP server.

For a detailed breakdown of the project goals, see .