Large language models (LLMs) are increasingly used in education to enhance learning experiences. However, challenges for LLMs in educational settings include concerns about the accuracy and appropriateness of AI-generated responses, particularly when addressing the literacy level needed for children. In this paper, we designed an elementary education math chatbot called MathBuddy to help children learn math more efficiently. MathBuddy serves as a learning assistant by offering children step-by-step guidance as they solve their math problems. We conducted a user study with 10 elementary school children (in grades 5 to 7) to evaluate the effectiveness of our chatbot. We found that children had an overall positive experience using the chatbot, as MathBuddy made learning math more fun and engaging. Our findings underscore the potential of the LLM-based chatbots to create a more interactive and effective math learning experience for children.