The catalytic cleavage of carbon−sulfur (C�S) double bonds on the metal sites without deactivation has aroused great interest in both fundamental catalytic research and industrial chemistry. Herein, activity descriptors are developed via machine learning and density functional theory (DFT) calculations to screen transition-metal single-site catalysts, which quantify the effect of both atomic electronic properties and coordination configuration on the hydrolysis of C�S double bonds. The valence electron number and electronegativity of active sites are found to be well related to C�S activation and sulfur poisoning, where Fe demonstrates high catalytic potential among a series of metal centers. On the other hand, the isolated Fe 1 and Fe 2 sites favor carbonyl sulfide (COS) adsorption and activation, while the COS easily dissociates into *S and *CO on Fe 3 hollow site, thus resulting in the formation of robust Fe−S bonds and catalyst deactivation. As anticipated, the as-designed Fe 1 −N 4 site achieves a COS conversion of ca. 96% at 100 °C, slightly better than the Fe 2 −N 4 site, approximately 8 times higher than that of the Fe/C, which is also better than those of other monatomic catalysts (such as Co-NC, Ni-NC, Sn-NC, and Bi-NC). The combination of in situ characterizations and theoretical calculations suggests that *COS and *H 2 O/*OH have a competitive adsorption relationship on Fe−N 4 sites, and two Fe−N 4 sites can synergistically catalyze the COS hydrolysis through the spilled H and OH.