Solid oxide fuel cell (SOFC) technology is of great potential for efficient combined heat and power (CHP) generation with low overall emissions. To ensure safe and efficient operations of μ-CHP SOFC systems, appropriate control and optimization methods are required. This is generally achieved by using model-based optimization techniques, whereby operating conditions that maximize efficiency and ensure constraint satisfaction are sought. However, mismatch or disturbances always exist between a ‘plant’ (actual system) and its model, and standard model-based techniques are thus not sufficient. To circumvent this, a constraint-adaptation (CA)-based real-time optimization (RTO) algorithm has been proposed to drive the system to optimal efficiency, while enforcing constraint satisfaction, despite plant-model mismatch and disturbances. The μ-CHP SOFC system, depicted in Figure 1, comprises six components: a catalytic partial oxidation reactor, a SOFC cell stack, an afterburner, a startup burner, an air heat exchanger, and a water heat exchanger. RTO does not require a high-accuracy model since it mostly relies on updating the model optimality conditions through the available measurements. In this context, a lumped model that captures the main features of the μ- CHP SOFC system is developed and tuned to match experimental data. The performance of CA-based RTO was systematically evaluated. Various scenarios aiming at optimizing either electrical or CHP efficiencies for different load profiles, were tested to verify the robustness of the proposed methodology. CA-RTO not only demonstrated its capability to optimize the SOFC system efficiently but also maintained the system within a safe operating window even despite disturbances and plant-model mismatch.