In DC microgrids, distributed generation unit (DGU), loads, and other components are tightly interconnected through power electronic converters and power lines.Once a local fault occurs, it can rapidly propagate, triggering cascading effects and significantly complicating state estimation and fault diagnosis.To address this challenge, this paper proposes amulti-agent-driven distributed optimal state estimation and fault localization scheme. By equipping each DGU with an agent capable of operating a distributed Kalman filter, each agent eliminates the adverse effects of interconnected subsystems based on a decoupled filtering strategy, thereby achieving optimal state estimation for local systems under the minimum variance criterion. Additionally, a distributed residual generator is constructed to enable the detection and localization of both single and concurrent faults. Finally, the effectiveness and superiority of the proposed method are validated through a DC microgrid system composed of three generation units.