This line [1] will become a source of many problems and wasted hours of debugging time. You could at least return a sample document based on random offset or scan, lets say 1000 docs and infer their shape. Mongodb is schemaless, so theres no guarantee all documents look the same. In fact the comment is inprecise > Handler for reading a collection's schema

[1] https://github.com/kiliczsh/mcp-mongo-server/blob/9ffcd8f9fe...