Don’t give AI your database.
Give it an interface.
Gentility connects Claude, ChatGPT, and other MCP clients to the queries, datasets, and commands you’ve approved — and nothing else. Every call logged. Every surface revocable.
Connecting AI straight to prod takes an afternoon. That’s the problem.
A community MCP server and your production credentials will genuinely work. Your team asks questions in plain English and gets real answers from real data. It feels like the future because it is. It’s also unbounded:
Egress is invisible.
Query results become model context. You can’t see what left, and you can’t get it back.
Attribution is gone.
“The AI did it” is not an audit trail.
Blast radius is total.
Credentials that can read everything, will.
Revocation is a project.
Turning it off means rotating credentials and hoping.
Publish what AI may use. Nothing else exists.
With Gentility you don’t grant access to a database. You publish specific surfaces to it:
- Queries
- Reviewed, parameterized, versioned. AI runs the SQL you approved, with the inputs you allowed.
- Datasets
- Shaped, PII-scrubbed slices of your data, kept fresh automatically. AI can query them freely, because you decided what’s in them.
- Commands
- Specific operations on specific servers, arguments pinned.
AI clients see the Catalog — the complete list of everything you’ve published. If it’s not in the Catalog, it doesn’t exist.
Removing a surface is one click, not a credential rotation.
Three steps.
Install the connector.
One package on one server. It links your databases and internal tools to Gentility.
$ gentility auth
Publish to the Catalog.
Write a query, shape a dataset, approve a command. Review changes the way you review code.
Connect your AI.
Claude, ChatGPT, Claude Code, or any MCP client. They get the Catalog — and a refusal for everything else.
Where your data actually goes.
Query results pass through Gentility’s service — that’s how MCP clients reach your connector. Here is exactly what that means:
- — Encrypted in transit, every hop.
- — Every query and command lands in your audit log — including its output, so you can see exactly what AI saw. Audit data is visible only to your organization.
- — We never train models on your data. Some features send content to OpenAI and Anthropic for processing only, under API terms that prohibit training on it.
- — Today, Gentility holds credentials for connected sources. A self-hosted mode that keeps credentials entirely inside your network is in development. We’ll say so here when it ships, and not before.
Under everything: policy and audit.
Every call passes policy before it runs — read-only defaults, blocked patterns, role scopes. Every call lands in the audit log, attributable to a person, a client, and a published surface.
Limits
Thresholds on rows, calls, and egress volume. Cross one and access stops until a human turns it back on. And there is always the kill switch.
Next: your codebase.
The same discipline — published, reviewed surfaces instead of raw access — applied to your repositories. Details when it’s ready.
Say yes — on your terms.
Free to start. Connect a database and publish your first query in under ten minutes.