The Capacitated Vehicle Routing Problem (CVRP) is a key logistics and computer science problem where a vehicle with limited capacity must deliver orders efficiently while minimizing costs. Traditional exact methods struggle with large-scale instances, making metaheuristic approaches like Tabu Search (TS) practical solutions to real-world problems. However, TS performance depends on usually manually tuned neighborhood search strategies. We propose a Tabu Search controlled by reinforcement learning (RL-TS) that dynamically selects the search policy for neighborhood move. Experimental results show that RL-TS improves the quality of the solution over standard TS in multiple CVRP instances and multiple problem sizes. This highlights the potential of machine learning-driven metaheuristics for complex optimization problems. Future work will refine the RL model and explore real-world applications.