We study the dynamics of clusters of active Brownian disks generated by motility-induced phase separation, by applying an algorithm that we devised to track cluster trajectories. We identify an aggregation mechanism that goes beyond Ostwald ripening but also yields a dynamic exponent characterizing the cluster growth z 1/4 3, in the timescales explored numerically. Clusters of mass M self-propel with enhanced diffusivity D - Pe2/ p . Their fast motion drives aggregation into large fractal structures, which are patchworks of diverse hexatic orders, and coexist with regular, orientationally uniform, smaller ones. To bring out the impact of activity, we perform a comparative study of a passive system that evidences major differences with the active case.