Abstract: Rapid flow-like landslide is a severe geological hazard, and accurately predicting their run-out process is essential for devising effective prevention and mitigation strategies. The material point method (MPM) is a promising numerical technique for simulating landslide dynamics, and has primarily been used for two-dimensional simulations along slope profiles. However, the two-dimensional simulations cannot reflect the influence of actual topography on the run-out process of landslides. Based on the open source MPM code MPM3D, this study developed a three-dimensional MPM model that incorporates complex sliding surface constraints. A simulation of a typical rapid flow-like landslide that occurred in Dagou Village, Tianshui City, Gansu Province, in July 2013, was conducted to examine its movement and deposition characteristics. The simulation results highlight that the Dagou landslide’s path and deposition were significantly influenced by the gully’s meandering terrain. During its run-out process, the landslide experienced several directional changes within the gully, going through four distinct phases: rapid acceleration, a sharp initial deceleration, a gradual slowdown, and a second significant speed reduction. The narrow throat area connected with the gully’s head acted as a constraint, causing most of the landslide mass to come to a stop at the slope’s base. A small portion, however, continued past this constriction, transforming into a debris flow that propagates along the gully until it rushes out at the gully mouth. The landslide’s peak velocity, 23 m/s, was observed in the transport zone. Even as it reached the gully mouth, the landslide’s leading edge maintained a speed of 6 to 8 m/s, resulting in substantial damage to structures at the mouth. The model’s predictions closely match the actual deposition area and morphology, validating the model’s capability to accurately depict the movement patterns of such landslides. This study advances our understanding of rapid flow-like landslide mechanisms and demonstrates the MPM model’s potential for more detailed landslide simulations.