Under standard assumptions, we show that the set of Markov Per- fect Equilibria in pure strategies is non-empty for stochastic repeated games with complementarities. We characterize the set of extremal values of those equilibria, and we show that those equilibria are unique fixed points of well-chosen operators. Those extremal equilibria can be approximated exponentially fast, and uniform convergence obtains for any initial guess chosen on a relevant functional set. This charac- terization also allows to generalize some standard results in monotone comparative statics known for one-shot games.