I'm really struggling here to wrap my head around this, so please bear with me...

From everything I read it seems like nonparametric permutational statistics in-and-of itself addresses the MCP. What I don't understand is why we would then apply the Bonferroni correction, cluster method, etc.

Using permutational stats we create a whole new distribution (histogram) and if the originally observed condition lies in the 5% tail of this new distribution we reject the null hypothesis. Boom, stats done... right?

As an aside: to explain my understanding, the MCP results in a nonparametric distribution because it assumes all the compared datapoints are independent, when in fact (in EEG) datapoints are related...? Is the interrelatedness irrelevant and the only focus of the MCP the increase in FWER due to the sheer number of statistical comparisons?

TL;DR: Why do we do both a nonparametric statistical test (permutation) and then Bonferroni or the cluster method?