This paper proposes Tool-Augmented Strategic Agent (TASA), a framework that leverages Large Language Models to improve solution quality for combinatorial problems. TASA decomposes the optimization process into traceable strategy selections, utilizing the information of the optimization trajectory to update solutions.