Real-world forecasting applications often require predicting a single target variable using multiple related time series as covariates. However, most existing models are designed for multivariate forecasting, i.e., jointly predicting all variables. This approach often fails to optimally leverage covariate information when only a single target series is of interest, resulting in unnecessary complexity and potential degradation in accuracy. We propose Frequency-Aligned Covariate Extraction Transformer (FACEformer), a novel framework for univariate time series forecasting with covariates. The FACEformer first identifies the most relevant covariate series in the frequency domain, selecting those with similar periodic characteristics to the target. It then extracts and reconstructs the dominant frequency components shared between the selected covariates, yielding frequency-aligned series. These frequency-aligned series, along with the target series, are fed into a specialized cross-variable Transformer-based architecture to learn inter-series interactions. By focusing on the target variable and its frequency-aligned covariates, FACEformer reduces unnecessary complexity and noise. Experimental results on real-world datasets demonstrate that FACEformer achieves improved accuracy over conventional multivariate models while offering better interpretability of covariate relevance. In summary, FACEformer provides an effective solution for univariate forecasting tasks by leveraging covariate information in a frequency-aligned and target-centric manner.