Ask a general-purpose assistant what’s wrong with your Sales database and it will give you a checklist: look for nulls, duplicates, orphaned keys. It can’t tell you which ones you have, because it has never seen the data.
The fix is to give it the facts. DataOps TestGen profiles every column in your SQL Server tables, generates and runs tests, and monitors freshness, volume, and schema. Its MCP server lets an assistant read all of that.
What you can ask
These map to tool groups on TestGen’s Supported Tools page:
| Question | What the assistant reads |
|---|---|
| ”What data do I have, and how healthy is it?” | Table groups, quality scores, and run summaries |
”What’s wrong with dbo.Orders?” | Hygiene findings and failed tests on that table |
| ”Show me the rows behind that failure” | The failing records, if your role and settings allow it |
| ”What changed since last week’s data profiling run?” | Profile differences between runs |
| ”Which tables are late or short today?” | Freshness and volume monitor results |
| ”Generate tests for the new table group and run them” | Test generation and test runs, under your role |
In a session DataKitchen published in July 2026, “What data do I have and how healthy is it?” returned a quality score of 78.9 on a neglected table group, and “What’s dragging the score?” surfaced 30 hygiene issues that no test covered. That demo used a different database, but the questions are the same on SQL Server.
How it fits together
- TestGen connects to SQL Server 2019 or later, Azure SQL Database, or Azure Synapse with a read-only login. SQL Server uses a SQL login. Azure SQL and Synapse also accept a managed identity or an Entra ID service principal.
- You run data profiling and generate tests. TestGen stores the computed results in its own database on a server you host.
- You add TestGen’s
/mcpURL to your assistant and sign in to TestGen in the browser. The setup guide has the steps, and our how-to covers the common connection errors. - The assistant calls TestGen’s tools. TestGen, not the assistant, holds the SQL Server connection.
Each tool runs with the signed-in user’s TestGen role. In Enterprise, the Business and Catalog roles are read-only and the Analyst role sees PII redacted.
What stays in your network
The MCP server is part of TestGen, and TestGen runs where you install it: on Windows next to SQL Server, on Linux, or in Kubernetes. A desktop or IDE client on your machine reaches it over your network. A hosted platform such as Databricks Genie calls it from its own cloud, so the server has to be reachable from there.
Your AI provider sees what the assistant reads from tool results. If that’s a problem for some data, TestGen Enterprise’s source data protection stops tools such as get_source_data and get_table_sample from returning rows and drops top values and extremes from the rest. The full breakdown is in what data leaves your environment.
Assistant alone vs assistant with TestGen
| Assistant alone | Assistant with TestGen MCP | |
|---|---|---|
| Knows what a column contains | Guesses from the name | Reads 55 profiled characteristics |
| Finds hygiene problems | Only in what you paste | 32 hygiene detectors on every data profiling run |
| Remembers last week | No | Run history, monitor trends, score history |
| Database access | Whatever login you give it | None; TestGen’s read-only login |
The MCP server is in open-source TestGen, free for one user, one connection, and one project. Enterprise is $100 per user per month plus $100 per connection per month. Start at the SQL Server install page.