This research delves into the problem of distributed adaptive failure compensation output feedback consensus for nonlinear multi-agent systems (MASs), specifically focusing on scenarios with multiple actuator failures that may exhibit unmatched redundancy, and directed switching communication graph. By leveraging estimates derived from neighboring agents, we design a novel distributed reference generator. To tackle the challenge of unmeasured state variables within each agent, we devise a reduced-order dynamic gain filter. Leveraging this generator and filter, alongside a recursive design approach, we construct a distributed adaptive protocol. This protocol employs adaptive techniques to effectively compensate for actuator failures, significantly loosening constraints on the communication graph, which can momentarily disconnect at any point in time instant. Furthermore, our approach reduces the number of introduced variables and the filter’s dimensionality, thereby mitigating numerical complexities. This work represents a pioneering effort in addressing output feedback consensus for nonlinear MASs, accounting for actuator failures and the possibility of unmatched actuator redundancy. Notably, the consensus error can converge to an arbitrarily small set, uninfluenced by actuator failures, ensuring global stability of the closed-loop system. Finally, simulation results verify the effectiveness of the proposed method.