I find that if I can't easily imagine how I would create a test for the code I'm currently banging on, I don't have enough encapsulation yet to arrive at a workable solution.
- will someone else be touching the code? If yes then I TDD
- Am I likely to be touching the code six months from now? If yes then I at least document with tests.
- Is it a crucial piece of code who's specs I want documented? If yes then I document with tests.
HOWEVER, if I have unit tests available, this is priceless, so if I can have the budget and time to write them, I would, but if writing tests would mean you avoid writing code, just write code (unless you work for SpaceX or NASA)
And to your point, if it's an MVP? I would say ditch the tests at first, it's better to test your idea and market, than test code that might be thrown away. and you should plan to throw away your MVP. you won't be able to live with yourself if you don't. just call it garbage code, and if you get traction, just rewrite the whole thing. it might be a lie your are telling yourself, but if you over engineer your startup, you might end up not knowing if it's true or not.
basically unit testing is on the function/class/module basis.
you break your design into modules, each having a clear role, and make them plug into each other in a way that you can "mock" the "rest of the world" (e.g. other objects) and test only that part.
to give an example, let's say you create a car, the unit test is to take a wheel, and spin it as if it's on the road, but without a car. take the lights, and plug them in as if they are in the car, take the engine and run it as if it's connected to the transmission, etc. in a car it's a bit more difficult, but in code, if your view needs a model, it doesn't matter if it was created from a database, a text file, or coded, as long as the interface fits.
There's a gap before when the point of TDD 'clicks' and you see the 'why'; before that it can seem a huge bore. After that it can seem a bore too in some situations but you can see the benefits of it.