Query your Postgres from ChatGPT. Without handing over your password.
ChatGPT and Claude in the browser can't run a local database server. BoltSchema gives them a hosted read-only endpoint — one URL, no config file, and a role that physically cannot write to your data.
Try it. No credentials. No catch.
One click provisions a hosted MCP endpoint backed by a freshly-seeded e-commerce database we host. Paste the URL into ChatGPT or Claude and ask it questions about real data — protected by the same read-only guard your own database will use.
Connect your AI client
http://localhost:3000/api/mcp/mcp_demo_…- 1.Open Settings → Connectors in ChatGPT or Claude.
- 2.Choose Add custom connector and paste the URL above.
- 3.Ask a question about your data.
No config file, no Node, nothing to install — browser clients speak this transport natively.
Database on localhost or in a VPC?
Hosted services can't reach it, and you shouldn't have to expose it to the internet to use AI with it. Run one command next to your database instead.
npx boltschema login
npx boltschema connectlogin opens your browser to approve a short code, so no token touches your shell history. connect asks for your connection string once and prints an endpoint URL to paste into ChatGPT or Claude.
Nothing exposed
The agent dials out over HTTPS. No inbound firewall rule, no open port, no public database.
We never see your password
The connection string stays on your machine. For agent endpoints we store no credentials at all.
Read-only, locally enforced
The agent re-checks every query itself and uses a READ ONLY transaction. It doesn't trust our server.
Works where local servers can't
ChatGPT and Claude in the browser accept a URL, not a local process. Paste one link — no config file, no Node, nothing to install.
We never see your password
Generate a scoped role in your browser and connect with that instead. Revoke it any time with DROP ROLE — your application's credentials never change.
Read-only at two layers
A SQL guard rejects mutations before they run, and every query executes inside a Postgres READ ONLY transaction — so even a guard miss is refused by the database.
Your database stays where it is. We hold a read-only role you create, scoped to exactly what you grant it, revocable in one query — and every request the AI makes runs inside a transaction that cannot write.
Supported Connections