Mixed crop-livestock systems are central to livelihoods in semi-arid West Africa, yet they face persistent challenges related to low input efficiency, nutrient depletion, and food insecurity. These systems are characterized by complex social and biophysical interactions that produce nonlinear outcomes, complicating sustainable management. To unravel this complexity, we developed TENGA-an agent-based model (ABM) simulating the dynamics of 700 heterogeneous farms in semi-arid Burkina Faso. The model captured spatial and temporal interactions among crop and livestock management, nutrient flows, and farmers' interactions. A Morris sensitivity analysis was used to identify key parameters influencing system behavior, and scenario analysis explored interventions to improve crop-livestock integration including legume intensification (increasing the functional contribution of nitrogen-fixing legumes), fertilizer subsidies, and intensified direct exchange of manure and crop residues. Nitrogen uptake efficiency, livestock feed parameters, manure management, and social interactions emerged as critical drivers of system outcomes. Sensitivity analysis revealed nonlinear feedbacks, particularly in the propagation of farm-level effects (e.g. income) to village-level dynamics (e.g. equity). Antagonistic responses among farm types reduced overall system gains, highlighting the interconnectedness of component interactions. Scenario results showed that while nitrogen-fixing legumes improved yields and reduced nutrient mining, they also reduced herd size and increased transhumance pressure. Fertilizer subsidies favored wealthier farms but did not address food insecurity among subsistence households. Intensified residue-manure exchanges enhanced nutrient recycling but had minimal impact at system level. This study demonstrates the utility of ABMs for exploring the multi-dimensional and multi-level dynamics of mixed farming systems of semi-arid West Africa.