Epilepsy affects around 50 million people worldwide, with approximately 30% of cases proving drug resistant. An alternative treatment for drug-resistant cases, called ketogenic diet (KD), leverages a carbohydrate poor diet, and has been pioneered in the early 20th century. We introduce a model of a population of quadratic integrate-and-fire neurons where the metabolic feedback associated with KD affects neuronal excitability. Our proposed mechanism for seizure control hinges on the presence of adenosine triphosphate (ATP)-dependent potassium channels, whose activity, in the absence of ATP, results in neuronal hyperpolarization. Through bifurcation analysis of a derived mean-field model we reveal the relation between the neurophysiologically desired and seizure-like states. The desired states correspond to equilibria of the mean-field model and are characterized by asynchronous low-rate neuronal activity, whereas the seizure-like states correspond to periodic solutions with synchronous local activity. We reveal two qualitatively different scenarios for the onset of synchrony. For weaker coupling, bistability between the lower- and the higher-activity asynchronous states unfolds from a cusp bifurcation, and the collective oscillations emerge via a supercritical Hopf bifurcation. For stronger coupling, one finds a complex scenario unfolding via a degenerate Bogdanov-Takens bifurcation, such that both lower- and higher-activity asynchronous states undergo transitions to collective oscillations, with hysteresis in the vicinity of subcritical Hopf bifurcations. We show that switching between the states can be controlled by perturbations of the ATP production rate, external stimulation, and pulse-like ATP shocks, and indicate a potential therapeutic advantage of hysteretic scenarios.