i'm working on a streamlit app where users can input natural language to query the database, (example "is the length of a hospital stay related to age"). database doesn't store the actual data, but the **metadata **of the table (column types, column descriptions etc) the json metadata of each table is stored as a record in a mongoDB database. attached a reference
currently stuck on the mcp client / server that uses gemini flash 2.0.
i was thinking of some tools like (1) get table names from mongoDB database (2) get the schema of a table, which is stored in a json (3) get relevant columns that can answer the question based on description etc.
something like this server attached but i am wholly stuck and chatgpt is clueless too
any kind of help will be good thank you i'm getting pretty desperate hahahah