Because DSLs are fun and MBeans aren't
The general idea here is you define some MBeans and their attributes to retrieve using a DSL, and jawnitor does just that. It then does the good dead of dumping them into a file somewhere (via slf4j perhaps?) so that your favorite monitoring tool, spread sheet editor, or scientific calculation library can pull them in and have their way with them.
But wait, aren't there like 100,000 tools out there that can read MBeans off of my server via remote JMX anyway? Yes, there are, but they need access to your machine, and need to be separately configured with each iteration of your app. By pluggin jawnitor into your app you don't need all that extra coordination and open ports and stuff.
Further down the road you can also manipulate these values, or add new ones before outputting them. This way you can do cool stuff like combine your metrics in some cool way into one super metric, or track the differential or integral term of some metric over time and base output on that.
Clearly this isn't done yet, but minimum viable product has been achieved, to some degree. From here standardization and hardening is the name of the game.