You get n as the derivative.
Which is always positive.
Thus the likelihood function is always increasing.
In order to maximise the likelihood function, make a the greatest possible.
But you know that
Since the xi's are observations, they're already realised. So find a with respect to xi. Such that it's the greatest possible.
Would you agree that ?
A similar reasoning is made for the MLE of a uniform distribution
I've seen that on a wikipedia (can't remember in which language), so you should find an example over the internet !