OK people, we are going to level up on Claude Desktop/Code usage this month. I am presuming that everyone following along here has a 16GB+ Mac, or they’re a fellow tinkerer with Linux, a GPU, and Claude Code, who can do for themselves on adapting things to their environment.
Attention Conservation Notice:
This will be paid subscribers only, get used to Paying To Play when it comes to AI. Sorry, that’s just how it is with AI, you either level up, or get left behind, and my list of $20/month subscriptions ain’t gettin’ any shorter.
Objectives:
Over the next thirty days we are going to do all of the following:
Install Claude Desktop.
Install Claude Code.
Installer Docker Desktop.
Create Claude Commands.
Create Claude Agents.
Create Claude Skills.
Connect Docker, JavaScript, and Python MCP servers.
Learn to evaluate MCP server quality based on its Github.
Store tabular data in Sqlite3.
Locally store and semantically search documents in Chroma.
Use the simple knowledge graph example MCP server.
Consider the differences between Claude, ChatGPT, and Perplexity.
Understand the differences between the Claude local and hosted solutions.
By the end of this month every one of you will have a Claude Desktop setup, probably with Claude Code installed as well. You’ll know when to create a Command, an Agent, or a Skill. You’ll understand the three methods of connecting MCP servers, and you’ll be able to tell if a server you find online is well maintained. We’re using Claude Code, but we are NOT programming. It’s just a slick way to do batch work that is absolutely infuriating using Claude Desktop.
We’ll get some tables in Sqlite3, we’ll get some documents into Parabeagle, and we’ll use the lighter, simpler file based knowledge graph reference server, rather than the poorly maintained chonker that I use, known as Memento. We’ll look at the massive list of available MCP servers, and if you’ve got a compelling case to use a particular one, we’ll drag it into the mix and make it go.
Claude, ChatGPT, and Perplexity all have different foundations. They have different strengths and weaknesses. They have different revenue models. I use the first two daily and I am extremely interested in Perplexity, but not sure if I’ll get time to properly evaluate it before we get to that point.
You are going to be using these things to do stuff for yourselves, but I think all of you have the urge to prototype stuff that will later need to scale up. That prototyping will be a Claude thing, not ChatGPT or Perplexity, and we’ll be sticking to MindsDB as the target for hosting. There are many options, that one happens to be FOSS - both gratis and libre, as well as being the one I chose to use as the foundation for various projects. And best of all, it’s a Python project.
We are going to lean heavily on Nate B. Jones as a source for this. His thinking on the industry is crystal clear and he offers ongoing educational segments on YouTube, Substack posts, and supporting material via Github. If you find something of his that’s extremely interesting, you can mention it and I’ll try to put it in context.
You’re going to listen to me on this rather than just following Nate, because it’s a focused, one month effort, a complete integrated package. Everything he does is good, but there are no boundaries there. I have no idea how he stays on top of everything all at once the way he does, but I’m so happy to be a new subscriber to his work.
Any questions? Just email neal@rauhauser.net and I’ll try to answer them.
