I agree with you for common CLIs that the model will natural have in its training set, but when building your own CLI it seems like the self-documenting nature and progressive disclosure of MCP would have a huge advantage.
With a cli you don’t have the prefix cost, but your agent also doesn’t know that the tool exists, what it’s for, how to use it, etc. once you give it that information, then you have the same issues with prefix cost, right? Changing tools does break the prefix cache, but changing tools is a thing you can do. You can’t change (model awareness of) CLI, and if you could then it’d also break the prefix cache, or you’d have tool instructions scattered throughout the context without structure.
Dumping to file is a harness thing, not a MCP thing. A harness will do the same with large terminal outputs.
Likewise setting up a new CLI can be complicated. I never see an agent try to install an mcp, but I also don’t typically see it try to install a new CLI.

I don’t see how mcp is any more a fixed set of tools than CLI is. It’s just presented to the agent in a structured way, which is important for discovery and instruction following.

But yeah ease of chaining is certainly useful wrt CLI. That’s a huge win for sure.

Can you elaborate on LLMs using CLI more effectively than MCP? Do you have any sense of how likely it is just your domain or the MCPs you’re using? Do you have any theories about why that may be?

That’s true to an extent, but I’d disagree wrt something like dockerhub.

I don’t really agree. If you have a tool to do a thing, share it in case someone else could benefit, just be open about what it is.

I’m generally anti-ai but this is kind of exactly what vibe coding is for.

Someone has a problem right now and there is no tool to fix their problem how they want it fixed, so they throw some shit together for personal use and maybe someone else can use it if they want idgaf.
Why would they maintain that? It does what they need it to, when they need it. Usually these tools are very basic.

I hate how so many of the arr apps don’t describe what they do in a way that people who don’t already know can understand.

Even the tutorials and guides are frustratingly vague.

I was just testing it out lol.
There is also a graph plugin for an obsidian-like view, but it’s kinda janky.

There is a beta AI plugin too, which (as much as I hate AI) is something I may try. The power of NotebookML is just too much, I wanna be able to ask it shit about my D&D campaign notes.

I was going to switch to trillium next, but in the end I decided not because it relies on a database even more than Joplin does.

I’m in the process of giving Silver Bullet a try. It seems to be pretty well designed. I don’t really like that folders are just cosmetic and not useful for navigation, but I like that it is open source, that the documents are saved in the file system, and that it’s self-hosted instead of synced.

I mean, you don’t HAVE to do any of that stuff in Windows, it’s just helps a bit.

I’m sure there are plenty of windows horror stories. But almost every Windows computer I’ve had in the last decade, both custom and OEM, has worked pretty well out of the box. And almost every Ubuntu computer I’ve had over the last decade has had problems that weren’t trivial to fix.

I like Linux, but when people compare these problems like they’re the same just are missing the point.