Multi-agent model is well-known to suit design of complex systems. This paradigm allows describing autonomous entities to interact together directly or through their environment. It is specially adapted to design 3D simulations taking into account spatial constraints on agents. In this work, we have designed a multi-agent model which adds a feature to the classical representation of agent: a body, encapsulating a physical model of the agent. We have applied this model to lipids and proteins belonging to the inner mitochondrial membrane, a biological membrane. Information provided by atomic structures is available through international databases and has been used to design a shape-based grain model for the agent body. We selected a model with three grains per molecule in which each grain is characterized by a type determining how they interact together and consequently the agent behaviors. Lipids and two kinds of protein structures have been described within this model allowing us to simulate their organization in membranes.