Artificial intelligence (AI) has significantly advanced drug discovery, yet identifying optimal hit and lead compounds remains challenging. Structure-based drug design (SBDD) has emerged as a promising paradigm, but the inherent data biases and ignorance of synthetic accessibility render SBDD models disconnected from practical drug discovery. We introduce Rag2Mol-G and Rag2Mol-R, two retrieval-augmented generation (RAG) methods for discovering small molecules fitting 3D pockets. These two methods involve searching for similar small molecules that are purchasable in the database based on the generated ones, or creating new molecules from those in the database that can fit into a 3D pocket. Our experiments indicate that Rag2Mol consistently generates candidates with high binding affinities and drug-likeness. Rag2Mol-R, in particular, offers broader chemical landscape coverage and superior targeting compared to advanced virtual screening models. Both methods successfully identified inhibitors for the challenging target PTPN2, previously deemed undruggable. Our extensible framework can integrate various SBDD approaches, and the code is available at: https://github.com/CQ-zhang-2016/Rag2Mol .