Résumé : With the increasing integration of large lauguage models (LLMs) in education, there is growing interest in using AI agents to support student learning in creative tasks. This study presents an interactive Mentor Agent system named Mentigo, which is designed to assist middle school students in the creative problem solving (CPS) process. We created a comprehensive dataset of real classroom interactions between students and mentors, which include the structured CPS task management, diverse guidance techniques, personalized feedback mechanisms. Based on this dataset, we create agentic workflow for the Mentigo system. The system's effectiveness was evaluated through a comparative experiment with 12 students and reviewed by five expert teachers. The Mentigo system demonstrated significant improvements in student engagement and creative outcomes. The findings provide design implications for leveraging LLMs to support CPS and offer insights into the application of AI mentor agents in educational contexts.