MCP Make Me Happy
MCP Make Me Happy
Empowering productivity with integrated AI tools
Recently, the topic of MCP has suddenly become extremely popular, and I guess the main reason is there are a lot of MCP servers in the eco-system.
In order to increase productivity, I’ve spent some time to test some MCP combinations. After all, computer resources are limited and I can’t use all MCPs at once, so I’ve picked a few that I find most useful.
I usually use the Smithery MCP registry, which has a large number of projects posted. Here are my favorite combinations.
First of all, sequential thinking I believe is the core, even though nowadays AI models usually have CoT, it is still necessary to ask AI to disassemble the workflow through prompt. By disassembling the workflow, the AI will be able to correctly invoke the corresponding tools to solve the problem. So sequential thinking plays the role of the brain.
Next, the desktop commander is the body that does all the work. Because the desktop commander can read files, write files, and execute commands, it can do everything a command line can do.
Finally, there is duckduckgo search, which provides additional information to accomplish specific tasks. I have to say this is not really used very often, but if we need to find the most recent information, a web search can be helpful.
Demo
This project is a weather prediction program I built to explain machine learning to team members. It took less than 10 minutes to complete without opening an IDE or writing a single line of code.
https://github.com/wirelessr/scikilearn-demo
The Claude Desktop and the three MCPs mentioned above were used.
System instruction: You are a professional ML developer who is familiar with various python ML frameworks and is able to call on various tools to actually operate the computer, e.g. read files, write files and execute commands.
We start by setting up the roles and game rules for the AI, and then we actually move on to development.
Help me create a scikit-learn demo project on my desktop to predict temperature, need to have test data. Also, this project needs to have a virtual env that does not pollute my system python.
At this point Claude will start building a draft of the project, but he has not separated train from predict and has not yet trained the model. Therefore, let’s ask the AI to modify the project and actually build the model.
Is this script already trained? How can I use it to predict tomorrow’s temperature? Run the commands directly to build the model.
Claude then creates the python environment via virtual env and installs the dependencies via pip. Once the environment is set up, Claude actually runs python src/train.py to train the model.
Go to the Internet and look up today’s weather information and use this project to predict tomorrow’s weather.
This is where duckduckgo comes in and looks up the required parameters, such as today’s temperature and humidity, and runs python src/predict.py to provide the final result.
At this point, the project is completed, but for ML teaching is not enough, so I then asked Claude to write a clear description and comments.
This is a tutorial project and I need README.md and good comments.
In the end, the project is what you see on Github. I didn’t write a single line of code and I didn’t use an IDE, but the project is complete.
Originally published on Medium