No API? No Problem.

Last July, I wrote about the AI tool that blew my mind. To be sure, model context protocols still let citizen developers do powerful things. Sadly, though, they waste oodles of tokens in the process. Even the company that invented MCPs (Anthropic) admits as much. (Some have equated the process to reading the encyclopedia before answering a question someone has asked you.)

In the ten months since I wrote that post, Claude has added oodles of features. What's more, the developer community has been busy. (Hell, even I released a neat public GitHub repo.)

One of the buzziest open-source projects today is Printing Press. Think of it as a library of bespoke command-line interfaces (CLIs) that anyone can use in myriad ways. (As its name implies, you can also roll your own.) In this post, I'll explain on way in which generative AI is breathing new life into this decidely unsexy UI.

Burning Basketball Questions

As my site's readers know by now, I'm an avid basketball fan. I'll often check ESPN NBA box scores to see which players performed well and which did not. (Wemby's numbers never cease to amaze me.)

Basic site searches are easy enough. If I wanted to ask more complicated basketball questions, though, I'd have to get creative. First, I'm no developer. Second, ESPN discontinued its API in December 2014.

Enter Printing Press.

I launched Claude Code and installed its ESPN CLI Skill. Here's my initial prompt to it requesting data on James Harden—the Cleveland Cavaliers' well-traveled starting point guard.