In a recent community call, Orbis Holding explored their implementation of a natural language Q&A system over a Memgraph dataset, which involved transitioning from a traditional LLM to Cypher pipeline to a more flexible Model Context Protocol (MCP) framework. The original approach, which translated user queries into Cypher queries, often failed due to mismatches in query structure and database schema, resulting in broken queries or empty results. The MCP framework, developed by Anthropic, offers a dynamic and iterative query generation process that allows for tool discovery, query refinement, and schema exploration, significantly improving the system's reliability. The shift to MCP increased the accuracy of query resolution from roughly 20% to 90% by allowing multiple reasoning cycles instead of single-shot execution, thereby providing deeper insights and more adaptable query handling. Challenges in transitioning included redefining workflow design and constructing MCP client logic to manage tool selection and iterative execution, but the new system promises greater scalability and accuracy for complex queries. Future improvements are aimed at reducing query completion time and enhancing tool selection efficiency.