<p>Persistent divalent metal cations such as Cu²⁺ and Mg²⁺ accumulate in freshwater systems and pose measurable risks to water quality and human exposure. Established analytical techniques, including inductively coupled plasma mass spectrometry and atomic absorption spectroscopy, require centralized laboratories, extensive sample preparation, and batch processing, which limits their use for continuous in situ monitoring at low concentrations. This study presents a terahertz frequency MXene-based metasurface sensor that integrates graphene, silver, strontium titanate, and copper layers to increase electromagnetic field confinement and coupling between the sensing surface and dissolved ions. The sensing mechanism relies on conductivity changes in MXene nanosheets induced by ion adsorption, modulation of the graphene surface response through electrostatic gating, and refractive index sensitivity in the terahertz band. Finite element simulations show that the sensor achieves a spectral sensitivity of 151.1&#xa0;GHz/RIU for Cu²⁺ and 227.8&#xa0;GHz/RIU for Mg²⁺. The corresponding figures of merit are 1145.5 and 1759.5 RIU⁻¹, with theoretical detection limits of 4.7 × 10⁻⁴ RIU and 2.7 × 10⁻⁴ RIU. Resonance frequency shifts exhibit linear relationships with refractive index changes, with <i>R</i>² values above 0.993, and with ion concentration, with <i>R</i>² values above 0.895. Random Forest regression was employed as a surrogate modeling tool to interpolate sensor responses obtained from physics-based simulations. The dataset was divided using an 80:20 train–test split, and model performance was evaluated on unseen test data. High coefficients of determination (<i>R</i>² &gt; 0.998) reflect the smooth, deterministic nature of the simulated electromagnetic response rather than experimental variability.</p>