The nitrogen reduction reaction (NRR) is a renewable alternative to the energy- and CO2-intensive Haber-Bosch NH3 synthesis process but is severely limited by the low activity and selectivity of studied electrocatalysts. The Chevrel phase Fe2Mo6S8 has a surface Fe-S-Mo coordination environment that mimics the nitrogenase FeMo-cofactor and was recently shown to provide state-of-the-art activity and selectivity for NRR. Here, we elucidate the previously unknown NRR mechanism on Fe2Mo6S8 via grand-canonical density functional theory (GC-DFT) that realistically models solvated and biased surfaces. Fe sites of Fe2Mo6S8 selectively stabilize the key *NNH intermediate via a narrow band of free-atom-like surface d-states that selectively hybridize with p-states of *NNH, which results in Fe sites breaking NRR scaling relationships. These sharp d-states arise from an Fe-S bond dissociation during N2 adsorption that mimics the mechanism of the nitrogenase FeMo-cofactor. Furthermore, we developed a new GC-DFT-based approach for calculating transition states as a function of bias (GC-NEB) and applied it to produce a microkinetic model for NRR at Fe2Mo6S8 that predicts high activity and selectivity, in close agreement with experiments. Our results suggest new design principles that may identify effective NRR electrocatalysts that minimize the barriers for *N2 protonation and *NH3 desorption and that may be broadly applied to the rational discovery of stable, multinary electrocatalysts for other reactions where narrow bands of surface d-states can be tuned to selectively stabilize key reaction intermediates and guide selectivity toward a target product. Furthermore, our results highlight the importance of using GC-DFT and GC-NEB to accurately model electrocatalytic reactions.