Intro to MCP Demo
The MCP Demo : Yr weather service using Model Context Protocol
The demo is a simple weather service that uses Yr web services to provide weather information.
What we will build
The demo is a web application where you can ask about the weather at a given place and get the response which will use a tool that uses Yr web services to get weather information. The demo demonstrates using MCP tools.
Tools is a category of capabilities that are offered in the MCP protocol.
Server features that are offered with MCP are:
- Tools: Executable functions that allow models to perform actions or retrieve information
- Resources: Structured data or content that provides additional context to the model
- Prompts: Predefined templates or instructions that guide language model interactions
The screenshot below shows the Demo in action, where I ask about the weather in Trondheim tonight. I specify it to include details also.
Screenshot of the demo
As the screenshot shows, we can ask free-form about the weather the AI LLM model which makes it possible to have a more natural conversation about the weather than just using a website to look up the weather.
Example of a more advanced demo would be: We could have combined this for example with Synthesized Text to Speech voice assistant and Speech to Text to have a natural conversation with the AI model.
The demo here focuses on the following Server Feature of MCP : Tools.
Yr is a popular weather service in Norway delivered by Norwegian Meteorological Institute (met.no) in Oslo, Norway.
In the same demo repo, you find code how you can expose metadata about the tools the MCP offers.
We can with for example Swagger present metadata information about the tools that are offered in the MCP protocol. This is a good way to present the contract of the MCP protocol and what capabilities it offers.
Github repo for the MCP Weather Demo
The Github repo is publicly available here for those who want to explore the code and try it out themselves:
https://github.com/toreaurstadboss/WeatherMCPDemo
Note that to run the demo, you must have an account at Anthropic web site and set up an API key here:
https://console.anthropic.com/dashboard
It is possible to pay as you use to test out Anthropic LLMs. I paid 20 dollars a year ago and still have not spent all of it. And this includes a lot of testing and playing around using the Anthropic Claude service in the demos I have written.
Prize $$$ vs complexity/performance of different Claude models
The model I have been using while testing is claude-3-haiku-20240307. It is some months since I worked most with the demo and I know newer Claude models are available now and suggested for those who wants to test out the demo themselves. There are newer models, but bear in mind that these Claude models are more expensive to use (3x).
Video of the demo in use
The following video shows the demo in use, where I ask about the weather in Trondheim tonight and get a response that includes details about the weather.
The video demonstrates how the demo works and how it uses the MCP protocol to provide weather information.