Significant advancements in semantic analysis for structured relational databases have led to numerous research models and commercial products in this domain. However, current efforts, particularly among large language models, often prioritize broader cross-domain semantic parsing tasks, frequently overlooking the specific requirements of specialized scenarios. This gap is particularly evident in database-focused research, such as Text-to-SQL tasks, where unique complexities and demands frequently arise. To address these challenges, this paper introduces M-SQL, a novel hybrid method designed for single-task scenarios with relatively fixed requirements. M-SQL effectively tackles the intricacies of generating SQL queries for complex, real-world Text-to-SQL tasks. The proposed method is rigorously evaluated on the Chase Text-to-SQL dataset, using 40 databases for experimentation. The results indicate that M-SQL significantly outperforms both state-of-the-art methods and leading large language models. By showcasing M-SQL’s exceptional capabilities, this work aims to provide valuable insights for industries reliant on efficient database queries, facilitating the rapid development of high-quality end-to-end database query interfaces. This contribution advances the field of semantic analysis for structured relational databases and opens new avenues for addressing complex Text-to-SQL challenges in real-world applications.