1. Tropical agroforests are diverse systems where several predator groups shape animal communities and plant-arthropod interactions. Ants, birds and bats in particular can reduce herbivore numbers and thereby increase crop yield. However, the relative importance of these groups, whether they interact, and how this interaction is affected by management and landscape context, is poorly understood. 2. We jointly manipulated access of ants, birds and bats in Indonesian smallholder cacao agroforestry across gradients of shade and distance to natural forest. We quantified arthropod abundance, pest damage and yield. 3. In control treatments, yield was highest under 30-40% canopy cover. Ant exclusion strongly reduced yield (from 600 to 300 kg ha À1 year À1 ) at 15% canopy cover. Bird exclusion impaired yield (from 400 to 250 kg ha À1 year À1 ) at 60% and enhanced yield (from 600 to 900 kg ha À1 year À1 ) at 15% canopy cover, while bats had no effect. Yield increased with forest proximity, a pattern not related to predator access. 4. No interactive effects among predator exclusions on yield, pest damage and arthropod communities were found. Ant exclusion increased numbers of herbivores below 30% canopy cover, without reducing spider abundances. Bird exclusion reduced herbivore and increased spider abundances. 5. Synthesis and applications. Using exclusion studies, we estimated that ants and birds cause cacao yield to vary between 100 and 800 kg ha À1 year À1 , depending on shade-tree management. In all but the most shaded agroforests, ants were pivotal in supporting yields. Yields under low-canopy cover were strongly dependent on access by predator groups, with birds reducing rather than increasing yield. Hence, cacao farmers should refrain from disturbing ant communities and maintain 30-40% shade-tree canopy cover not only for ecophysiological reasons but also to buffer variability in predator communities.