This study addresses methodological and contextual gaps in carbon stock estimation by developing a UAV-based model tailored for cocoa agroforestry systems in tropical smallholder landscapes. As nature-based solutions gain traction in climate change mitigation, reliable monitoring of agroforestry systems, marked by spatial heterogeneity and farmer-managed canopies, becomes increasingly vital. The research aims to develop a biomass estimation model based on RGB sensor-derived UAV imagery and a modified allometric equation. A total of 35 cocoa agroforestry plots were selected in Luwu Timur, South Sulawesi, based on strict biophysical and accessibility criteria. RGB imagery collected via DJI Phantom 4 UAVs was used to generate Canopy Height Models (CHMs), which were validated against field measurements of tree height and diameter. A modified allometric equation incorporating both variables were used to estimate above-ground biomass (AGB), which was subsequently converted into carbon stock values. The UAV-estimated tree heights demonstrated strong correlation with field observations (R² = 0.7344), confirming model reliability. This technique has proven capable of substituting for tree height estimation using LiDAR, which requires more advanced and costly equipment. Estimated carbon stocks ranged from 18.41 to 134.49 tCO₂e/ha, highlighting the variability across agroforestry systems shaped by diverse management practices. This study presents a replicable and scalable framework for integrating UAV-based methods into carbon finance schemes.