An MCP tool schema is not just documentation. It is part of the model’s operating environment. The model reads the tool name, description, input schema, and output shape to decide: whether to use the tool what arguments to send what the result means whether the tool is safe for the user’s intent That means schema drift can become behavior drift. Why this matters for database tools If a normal API changes, a typed client or test suite may fail. If an MCP database tool changes, the agent might keep running but behave differently: a parameter changes and the agent stops using the tool a description becomes broader and the model overuses it an output field disappears and the agent reasons from missing context a tool name hides whether it is read-only or write-capable For database-connected agents, the schema is part of the safety boundary.…