Modular multilevel converters provide high power quality, high reliability and flexible voltage and power scalability. However, large capacitor voltage ripple is one of the main drawbacks of this topology. This ripple reduces the capacitor lifetime and, due to the large capacitance required to limit it, increases the size and cost of the converter. In this work, the capacitor voltage ripple is expressed in terms of energy and decomposed into components which have a defined frequency harmonic behaviour. Using this decomposition, a control scheme to eliminate the low frequency components and, hence, to minimize the amplitude of the voltage ripple is proposed. The proposed control scheme generates references for circulating current and common-mode voltage achieving a reduction of 60% of the original ripple.