Reconciling agricultural production with climate-change mitigation and adaptation is one of the most formidable problems in sustainability. One proposed strategy for addressing this problem is the judicious retention of trees in agricultural systems. However, the magnitude of the current and future-potential benefit that trees contribute remains uncertain, particularly in the agricultural sector where trees can also limit production. Here we help to resolve these issues across a West African region responsible for producing ≈60% of the world's cocoa, a crop that contributes one of the highest per unit carbon footprints of all foods. We use machine learning to generate spatially-explicit estimates of shade-tree cover and carbon stocks across the region. We find that existing shade-tree cover is low, and not spatially aligned with climate threat. But we also find enormous unrealized potential for the sector to counterbalance a large proportion of their high carbon footprint annually, without threatening production. Our methods can be applied to other globally significant commodities that can be grown in agroforests, and align with accounting requirements of carbon markets, and emerging legislative requirements for sustainability reporting.