An agent-based model (ABM) is developed to simulate the impacts on streamflow and groundwater levels by the dramatic increase of hydraulic fracturing (HF) water use. To develop the agent-based model, institution theory is used to model the regulation policies, while evolutionary programming allows agents to select appropriate strategies when applying for potential water use permits. Cognitive maps endow agents’ ability and willingness to compete for more water sales. All agents have their influence boundaries that restrict their competitive behavior toward their neighbors but not to non-neighboring agents. The decision-making process is constructed and parameterized with both quantitative and qualitative information. By linking institution theory, evolutionary programming, and cognitive maps, our approach is a new exploration of modeling the dynamics of coupled human-natural systems (CHNS) to address the high complexity of the decision-making process involved in the CHNS. The ABM is calibrated with HF water-use data, and the calibration results show that it is reliable in simulating water depot number, depot locations, and depot water uses. The SWAT (Soil and Water Assessment Tool) model of the Little Muddy River basin and the MODFLOW of the Fox Hill-Hell Creek regional aquifer are coupled with the ABM to simulate the changes in streamflow and groundwater level, respectively, under different scenarios such as HF water demand, climate, and regulatory policies. The integrated modeling framework of ABM, SWAT, and MODFLOW can be used to support making scientifically sound policies in water allocation and management for hydraulic fracturing.