Relational databases are valuable across many domains, but developers who are new to a system may find organically grown relational schemas complex to explore due to numerous tables linked transitively by their foreign-key relationships. While Large Language Models (LLMs) excel in natural language processing tasks, integrating them with external data requires customized accessors, which necessitates significant engineering effort to adapt to specific relational schemas and settings. Recently, the Model Context Protocol (MCP) has provided a way for LLMs to use external tools to take actions or retrieve data independently, instead of functioning solely as text generators. This delegation allows LLMs to interpret and fulfill user requests dynamically, shifting away from a fixed developer plan. We present a proof-of-concept MCP-based system for exploring relational data, confirming that an LLM can respond to user queries by generating and executing query-specific SQL, integrating a tool for discovering relational paths between tables using a graph of transitively connected tables.