Prepare the connection deliberately
Identify the database, the account or credentials it should use, and the records the task needs. Use an account with only the permissions required for that work rather than a broad administrative identity. Enter secrets in the connection settings, not in a prompt or a workflow description that may be shared or retained.
Confirm the connection points to the intended database and environment. A development database and a production database may have similar names but very different consequences. If the connection form accepts a database name or path, verify that value carefully; a connection can succeed while selecting an empty or unintended database.
Test access before relying on it
After saving the connection, run a small read-only query against a known collection or table. This checks more than whether the settings were accepted: it confirms that the credentials can reach the expected data. If the result is empty, verify the target and query before assuming that the records are missing.
Once access works, explain the data structure and business meaning to the agent or workflow that will use it. Review any proposed writes separately and keep a record of which connection is used in each automation. This makes it easier to troubleshoot permissions and helps prevent a test task from accidentally using a production dataset.