Cytokine storm is a life-threatening inflammatory response characterized by hyperactivation of the immune system. It can be caused by various therapies, auto-immune conditions, or pathogens, such as respiratory syndrome coronavirus 2 (SARS-CoV-2) which causes coronavirus disease COVID-19. Here we propose a conceptual mathematical model describing the phenomenology of cytokine-immune interactions when a tumor is treated by an exogenous immune cell agonist which has the potential to cause a cytokine storm, such as CAR T cell therapy. Numerical simulations reveal that as a function of just two model parameters, the same drug dose and regimen could result in one of four outcomes: treatment success without a storm, treatment success with a storm, treatment failure without a storm, and treatment failure with a storm. We then explore a scenario in which tumor control is accompanied by a storm and ask if it is possible to modulate the duration and frequency of drug administration (without changing the cumulative dose) in order to preserve efficacy while preventing the storm. Simulations reveal existence of a "sweet spot" in protocol space (number versus spacing of doses) for which tumor control is achieved without inducing a cytokine storm. This theoretical model, which contains a number of parameters that can be estimated experimentally, contributes to our understanding of what triggers a cytokine storm, and how the likelihood of its occurrence can be mitigated.