A governed lakehouse your team can simply ask questions ofเลคเฮาส์ที่กำกับดูแลได้ ซึ่งทีมของคุณ
ถามคำถามได้ตรง ๆ

A regional planner wants to know why Vietnam is behind plan. Volume sits in the sales system, parts availability in the supply chain system, machine health on the plant floor, and the approved budget in a workbook owned by Finance. Four teams, four extracts, and by the time the answer arrives the month has closed.

This demonstration keeps the governance and removes the wait. Each business domain owns and publishes its own data product. The lakehouse serves them from one place. A question asked in plain language reaches the governed data directly, and when the answer needs the financial plan as well, it blends the two without anyone exporting a spreadsheet.

This demonstration is not really about one tool talking to one database. It is about the platform publishing a single governed endpoint that any AI tool can consume. Amazon Quick Suite is the client wired up here and the one the demo drives, but nothing in the design is specific to it.

That is what the Model Context Protocol buys. The lakehouse offers six named tools rather than a connection string, so a client never handles credentials and never writes its own SQL against the raw schema. Authentication, tool discovery and the read-only guardrails all sit in one place, which means connecting a second tool changes nothing in the platform.

Three of these were the starting point: Backorder, Sales Performance and Machine Sensor. Two more were added because the interesting questions cross domains. Profitability and Cost turns an operational problem into a number the business recognises, and Service and Warranty is the bridge that carries a machine fault through to a warranty claim.

This is a real session, captured as it ran. Amazon Quick on desktop registers the MCP server itself rather than through a Quick Suite connector, so the lakehouse is reached by a local stdio proxy that mints and refreshes the OAuth token on its own. Nothing about the governance changes: the same seven tools, the same read-only database user, the same refusal of anything that is not a single SELECT.

What follows is worth watching for one reason. Nobody pasted a schema, a connection string or a credential into the chat. The agent discovered what was published, worked out which parts of the answer the lakehouse could not supply, and said so.