MVPs fail most often from scope creep disguised as 'essential features.' The discipline of MVP building is less about what to include and more about what to defer.
We see three common traps: the 'while we're at it' trap (adding features because development is already happening), the 'users will expect' trap (assuming requirements without validation), and the 'future-proofing' trap (building for scale before proving value).
The antidote is relentless focus on the core hypothesis. Every feature should connect directly to something you're trying to learn. If a feature doesn't help validate or invalidate your assumptions, it doesn't belong in the MVP.
A useful exercise: for each proposed feature, articulate the specific learning it enables. If you can't, the feature is probably premature.