Ask Claude or ChatGPT about the data in SQL Server or Azure SQL, limited to the tables you approve, with writes only where you allow them.
Read from SQL Server when someone asks. Nothing exported, nothing copied, and changes only where you allow them.
What were sales by branch last month, and which branch missed its target?
Which purchase orders are still open after 30 days, and with which suppliers?
Mark invoice 55120 as paid and set the payment date to today.
Each is its own permission, never on the User role. Stored procedures, schema changes, SET and explicit transactions are refused: every call is one auto-committed statement. Every write is on record.
Already on dbt? We import the project (dbt Core is enough), and its definitions guide every answer from SQL Server.
Years of operational data sit in SQL Server, often behind an ERP or an in-house app, and getting a report out of it means a ticket to IT. Connected once, your team asks in plain English and gets answers from the tables you approved, inside that one database and nothing else on the server.
An admin connects SQL Server one time. Nobody hands over a personal login.
Each answer is read at the moment of the question. Nothing is exported or copied, so there is nothing to keep in sync.
What each role can see and do holds in every AI your team uses, and in your own agents.
Every question and every action, by every person and agent, in one audit trail.
Yes, with SQL authentication: a username and password. Windows (NTLM) and Microsoft Entra ID authentication are not supported.
No. EXEC, xp_ and sp_ procedures, schema changes and explicit transactions are refused in every call. Reads are single SELECT or WITH statements; writes, if you turn them on, are single INSERT, UPDATE, DELETE or MERGE statements on approved tables.
No. The connected database is its whole world: USE, four-part names and linked-server functions are blocked in every query. Inside that database it sees only the tables you approve.
Yes, and one built for a whole company. FrontierOps gives Claude, ChatGPT and any other MCP client one connection to SQL Server, set up once by an admin. What each role can read or change is decided in FrontierOps, and every call is on record, instead of each person running their own server with their own login.
A direct connection links one AI to SQL Server through one person’s login, with that person’s settings. That works for one person. For a team it means no shared definitions, no single place to decide who can see or change what, and starting over when you switch AI. Through FrontierOps, SQL Server is connected once, for every AI and every agent, under one set of permissions, with every call on record.