MASON is a fast, easily extensible, discrete-event multi-agent simulation toolkit written in Java. It is designed to serve a wide range of multi-agent simulation tasks, from swarm robotics to machine learning and social complexity environments. MASON emphasizes model and visualization separation, allowing models to be dynamically detached or attached to visualizers and to change platforms mid-run. The system is open-source and free, developed by the Department of Computer Science and the Center for Social Complexity at George Mason University. MASON's architecture consists of three layers: the utility layer, the model layer, and the visualization layer. The utility layer provides essential functionalities such as random number generation, data structures, GUI widgets, and movie/snapshot generation. The model layer includes a discrete-event schedule, a random number generator, and fields for objects and values. The visualization layer supports GUI-based visualization and manipulation of the model, allowing for dynamic detachment and reattachment of the model from the visualization. MASON's design goals include a small, fast, and easily modified core, with no built-in features specific to social agents or robotics simulators. It aims to be a general-purpose simulator suitable for various research areas, including robotics, machine learning, and social systems. MASON supports 2D and 3D fields, sparse and continuous spaces, networks, and various visualization options. The paper compares MASON to other multi-agent simulation environments, highlighting its unique features such as dynamic model-visualization separation, platform independence, and speed. It also discusses six applications of MASON, including network intrusion, urban traffic simulation, cooperative target observation, ant foraging, anthrax propagation, and a model of memory and primitive social behavior in wetlands. These applications demonstrate MASON's broad applicability and utility in different research domains.MASON is a fast, easily extensible, discrete-event multi-agent simulation toolkit written in Java. It is designed to serve a wide range of multi-agent simulation tasks, from swarm robotics to machine learning and social complexity environments. MASON emphasizes model and visualization separation, allowing models to be dynamically detached or attached to visualizers and to change platforms mid-run. The system is open-source and free, developed by the Department of Computer Science and the Center for Social Complexity at George Mason University. MASON's architecture consists of three layers: the utility layer, the model layer, and the visualization layer. The utility layer provides essential functionalities such as random number generation, data structures, GUI widgets, and movie/snapshot generation. The model layer includes a discrete-event schedule, a random number generator, and fields for objects and values. The visualization layer supports GUI-based visualization and manipulation of the model, allowing for dynamic detachment and reattachment of the model from the visualization. MASON's design goals include a small, fast, and easily modified core, with no built-in features specific to social agents or robotics simulators. It aims to be a general-purpose simulator suitable for various research areas, including robotics, machine learning, and social systems. MASON supports 2D and 3D fields, sparse and continuous spaces, networks, and various visualization options. The paper compares MASON to other multi-agent simulation environments, highlighting its unique features such as dynamic model-visualization separation, platform independence, and speed. It also discusses six applications of MASON, including network intrusion, urban traffic simulation, cooperative target observation, ant foraging, anthrax propagation, and a model of memory and primitive social behavior in wetlands. These applications demonstrate MASON's broad applicability and utility in different research domains.