Set up with your AI agent

Hand the two code-touching setup steps — the read-only DB role and the Stripe webhook — to Claude Code or Cursor.

Two onboarding steps touch your own systems: creating a read-only database role, and registering a Stripe webhook endpoint. Hand both to your AI coding agent instead of doing them by hand.

Before you start

An AI coding agent (Claude Code or Cursor) open in your app’s repository, with permission to run SQL against your database and edit code.
Admin role on the ParityRail project, and access to your Stripe Dashboard — the agent hands back credentials that you paste into the ParityRail form yourself; it never touches ParityRail directly.
You’ve reached step 2 (Stripe) and step 3 (Database) of onboarding — the Set up with your AI agent card lives inside those two steps.

The flow

In ParityRail onboarding step 2 (Stripe), expand the Set up with your AI agent card (the sparkles row) and click the copy button. This fetches a Stripe-specific prompt containing your exact webhook URL and the restricted-key instructions.
Paste that prompt into Claude Code or Cursor in your app’s repository. Let the agent create the restricted key and register the webhook endpoint in Stripe.
In ParityRail onboarding step 3 (Database), on the Postgres / Supabase tab, expand the separate Set up with your AI agent card and copy that prompt — it contains the least-privilege read-only role SQL tailored to your introspected schema, plus the optional repair role. This card is not shown on the Clerk tab — Clerk needs no role SQL.
Let the agent run the SQL and assemble a read-only connection string. The prompt’s closing line tells it to report the connection string / key and signing secret back to you, and not to paste secrets anywhere except the ParityRail form.
Back in ParityRail, paste the restricted key into step 2’s API key field and click Test & connect; paste the connection string into step 3’s Read-only connection string field and click Test connection.
Finish the mapping steps and run the dry run (step 7).

Review what your agent proposes before it runs anything against production — especially role creation and grants. Both prompts ask for least privilege; keep it that way.

Nothing to hand-write

Both prompts are generated for you, inside the wizard — there’s nothing to author yourself. The Stripe prompt embeds your project’s webhook URL, /api/webhooks/stripe/<your-project-id>; the database prompt embeds the read-only role SQL. ParityRail also publishes these docs in a format your agent can read directly, without you pasting anything in:

https://<your-host>/llms.txt

A linked index of every doc page.

https://<your-host>/llms-full.txt

The full documentation as plain markdown, in one file.

Verify it worked

The generated prompt renders inside the expanded card in a scrollable mono block with a copy button and the caption “Paste into Claude Code, Cursor, or your agent of choice.” That’s the only signal the AI card itself gives you — the real success signals show up after you paste the credentials back into ParityRail:

Step 2 shows the green Stripe connected card plus the Key permissions checklist, with green checks for Customers, Subscriptions, and Invoices.
Step 3 shows the green Database connected card with the host and database name.

Troubleshooting

Symptom	Cause	Fix
The card spins, then shows an error instead of a prompt.	Prompt generation failed — usually Stripe or your database wasn't reachable for introspection.	Collapse and re-expand the card to retry; the fetch is re-armed on the next open.
Database prompt is missing on step 3.	You're on the Clerk tab. The AI-agent card only renders for the Postgres/Supabase source.	Switch to the Postgres / Supabase tab — Clerk needs no role SQL, so it has no AI-agent card. See connect Clerk.
Agent proposes a superuser grant or GRANT ALL.	The agent went broader than the prompt asked for.	Reject it. Both prompts ask for least privilege — SELECT only, plus column-scoped UPDATE for repairs — before running anything against production.
A secret ended up in a chat log.	The agent (or you) pasted a credential somewhere other than the ParityRail form.	Rotate it immediately. The prompt explicitly instructs the agent not to paste the connection string, key, or signing secret anywhere except the ParityRail form.

Operate ParityRail from your agent

Want your agent to operate ParityRail directly, not just read about it? Connect it over MCP (Model Context Protocol).

From there, it can list access incidents, inspect a customer’s state, and trigger a check. See MCP for AI agents for the full setup.