Bacteria exhibit a diverse range of ecological interactions dictated by the strategic allocation of environmental resources. While genome-scale metabolic models can identify these interactions, their high computational cost often challenges large-scale dynamic analyses. We present the Metabolic Bacterial Game (MetabGame), a computationally tractable framework that models bacterial behavior through the lens of strategic resource consumption and production. By generalizing multi-agent reinforcement learning to microbial ecology, MetabGame simulates how agents navigate their metabolic needs based on environmental availability. The framework is designed to operate under both perfect and imperfect information (Bayesian game), allowing for the modeling of agents that either fully observe or must infer the metabolic states of their partners and contenders. We demonstrate that MetabGame provides an efficient framework for modeling complex microbial systems, maintaining predictive accuracy while scaling to populations up to hundreds of agents.