Mining marine trajectory data has broad applications, e.g., offering valuable insights for individual navigation. Existing trajectory recovery algorithms often overlook the hidden features behind uncertain marine trajectories. This limitation hampers the accurate recovery of trajectories under low sampling rates. In this study, we introduce STAR, a Spatio-Temporal trAjectory Recovery system designed to recover real trajectories with limited information. STAR employs an integrated encoding module to capture correlations among temporal, spatial, directional, and velocity features, uncovering latent patterns in uncertain trajectory data. The system compares predicted trajectories against actual trajectories, providing a visual representation of recovering performance. Experimental results demonstrate that STAR improves the root mean square error (RMSE) by 3.74% compared to state-of-the-art methods, highlighting its effectiveness in trajectory recovery. The demonstration video is available at https://github.com/linng12145/STAR