<p>This paper investigates practical finite-time leader–follower consensus in output-constrained second-order nonlinear multi-agent systems (MASs) characterized by unknown dynamics and external disturbances without prior knowledge of the disturbances’ upper bounds. The problem is initially addressed through the construction of a modified fixed-time distributed observer, designed to estimate the states of the leader for each follower. Subsequently, a nonlinear transformation is applied to convert a system with the output constraint into an unconstrained system. Radial basis function neural networks (RBFNNs) are employed to approximate the unknown dynamics of both the leader and follower agents. An adaptive sliding mode technique is then utilized to overcome the estimation errors caused by the RBFNNs and to handle unknown external disturbances. The employed approach in designing the consensus protocol results in chattering-free control input. The proposed structure demonstrates its capability to achieve practical leader–follower finite-time consensus among agents in the presence of unknown dynamics, unknown disturbances, and output constraints. Finally, the proposed distributed observer and consensus protocol are simulated, and the results are compared with the previous studies.</p>