How do you work with MVP's within the context of Agile?
Let's start with some definitions; Eric Ries (the person who coined the phrase) describes MVP as "a version of a new product which allows a team to collect the maximum amount of validated learnings about customers with the least effort."
As per the illustration above, it sits snuggly in between being "viable" and "minimum". If you make it too "minimum", you get something your customer can't tell if they like it, hence not collecting the maximum validated learning. On the other hand, if you make it super "viable" (or so you assume), you spend lots of money and time building on assumptions before validating any. So the sweet spot is somewhere in the middle - minimum enough to not break your bank or time and viable enough your customers can validate your assumptions.
You also need to be aware of the mistakes usually made when calling something an MVP.