The development of catalysts for electrochemical reduction of carbon dioxide (eCO2RR) with high activity and selectivity remains a grand challenge to render the technology useable. As promising candidates, metal–nitrogen–carbon (MNC) catalysts with metal atoms present as atomically dispersed metal–Nx moieties (MNx, M = Mn, Fe, Co, Ni, and Cu) were investigated as model catalysts. The distinct activity for CO formation observed along the series of catalysts is attributed to the nature of the transition metal in MNx moieties because of otherwise similar composition, structure, and morphology of the carbon matrix. We identify a volcano trend between their activity toward CO formation and the nature of the transition metal in MNx sites, with Fe and/or Co at the top of the volcano, depending on the electrochemical potential. Regarding selectivity, FeNC, NiNC, and MnNC had Faradaic efficiency for CO >80%. To correctly model the active sites in operando conditions, experimental operando X-ray absorption near edge structure spectroscopy was performed to follow changes in the metal oxidation state with electrochemical potential. Co and Mn did not change the oxidation state with potential, whereas Fe and Ni were partially reduced and Cu largely reduced to Cu(0). Computational models then led to the identification of M2+N4–H2O as the most active centers in FeNC and CoNC, whereas Ni1+N4 was predicted as the most active one in NiNC at the considered potentials of −0.5 and −0.6 V versus the reversible hydrogen electrode. The experimental activity and selectivity could be rationalized from our density functional theory results, identifying in particular the difference between the binding energies for CO2*– and H* as a descriptor of selectivity toward CO. This in-depth understanding of the activity and selectivity based on the speciation of the metals for eCO2RR over atomically dispersed MNx sites provides guidelines for the rational design of MNC catalysts toward eCO2RR for their application in high-performance devices.