Terminology and intended audience for MCP

[olaservo]

Pre-submission Checklist

• I have verified that this discussion would not be more appropriate as an issue in a specific repository
• I have searched existing discussions to avoid duplicates

Discussion Topic

Hi! I've been excited about MCP since it first got announced. Plugging literally anything into Claude has been a lot of fun. Even more importantly, it feels like a gateway to building out a personal cognitive architecture, one that can actually support whatever niche interest or ideas I throw at it. To that end, I've been learning how to build a specialized client to better visualize and prototype different ways of connecting things together. (If anyone here uses Max/MSP, it's loosely inspired by it.)

Throughout this process I've had a hard time penetrating some of the terminology. One example is the use of the term "sampling" to describe content completion or generation by a model. As someone who works in a gray area between software engineering and machine learning, the word itself is familiar. And once I take the time to think about why it would be called that, it makes sense.

However I'm curious why a more 'user-friendly' term wasn't used. And this leads me to a bigger question: is MCP supposed to be something that most tech-builders can grasp? As it is, its not exactly approachable. Clients and servers are mostly clear at this point, but beyond that I needed to tag in Claude to help me out.

[jspahrsummers]

It's a good question! We mostly think of MCP as something that implementors (of would-be clients and servers) should know about, and something that hackers might want to know about, but mostly not as something end-user facing.

That's no excuse for confusing terminology, though, so we're open to feedback on any terms that are unclear. "Sampling" was a tricky one in particular because we already use "completion" to refer to argument completion. It might be difficult to change that now, but we should be cognizant of stuff like this going forward.