--help on your CLI is all you need for LLM context until you don't.

Here's something I've been meaning to write up for a while. It's kind of late, but I'm going to lean into the whole MCP vs. CLI argument because I have a perspective others might find interesting.What's the actual difference between an MCP server and a regular old CLI program with a `--help` flag? Feels to me like a CLI program written with agents in mind solves the same problems without inventing anything new?https://x.com/adamwathan/status/1968738637358522629Developer tooling companies can be sorted into two tiers: S-tier (winning) and F-tier (losing). The Rubik for wherever a company falls into the S tier or F tier has got everything to do with the model weights.An S-tier company is in the training weights. Let's take GitHub, for example. Now, we can yak about problems with GitHub till the cow comes home, but one thing is true: these models know how to interface with GitHub because that behavior is in the model weights themselves. They don't need any skills or an MCP to get…