Breast cancer is a leading cause of mortality among women, requiring early detection and diagnosis. Researchers are developing increasingly sophisticated AI models to predict breast cancer risk, assisting radiologists in screening. However, to learn effectively, the models require large amounts of data that can be difficult to source from a single location. Although there are various public mammogram breast cancer datasets, they differ in terms of data quality and format and cover diverse demographic, geographic, and outcome distributions. Thus, including them in a new AI model development is a challenge. In this study, we introduce Mammo-Find, an LLM agent-based tool for the discovery of public mammogram datasets using a multichannel user interface (text and visual). Mammo-Find has knowledge of 22 mammogram datasets and uses different LLMs and Retrieval-Augmented Generation (RAG) techniques. LLM responses are given as text and as a knowledge graph. This study highlights the feasibility of LLMs as agents for interacting with datasets, which can help reduce search time and ease the work of medical researchers.