Clerk says rate limit exceeded. Here is why, and the fix
Clerk's own docs describe this plainly: a 429 means you are rate limited, and further requests to that specific endpoint are blocked until a Retry-After window passes. The general Backend API limit is high (1000 requests per 10 seconds on production), but sign-in and sign-up attempts are Frontend API calls with a much lower, per-IP ceiling, and those are usually what a script, a shared office network, or a burst of real users actually hits.
Last checked November 18, 2026.
What the error means
Clerk's system-limits documentation states it directly: on a 429, "all subsequent requests to that specific endpoint will be blocked for a given amount of time," and the response carries a Retry-After header with the number of seconds until the block expires. The two APIs are keyed differently: "Backend API requests are rate-limited per application instance which is identified by the Secret Key," while "Frontend API requests are rate-limited per user and identified by their IP address" — which is why a shared office or school network, not just one bad client, can trip the sign-in/sign-up limits below.
Sources: Clerk: System limits
The documented limits
| Scope | API, keyed by | Limit |
|---|---|---|
| Backend API, general (production) | Backend API, per Secret Key | 1000 requests per 10 seconds |
| Backend API, general (development) | Backend API, per Secret Key | 100 requests per 10 seconds |
| Sign-in creation / sign-up creation | Frontend API, per IP | 5 requests per 10 seconds |
| Sign-in attempts / sign-up verification attempts | Frontend API, per IP | 3 requests per 10 seconds |
| Invitations | Backend API, per Secret Key | 100 requests per hour |
| Bulk invitations | Backend API, per Secret Key | 25 requests per hour |
| Organization invitations | Backend API, per Secret Key | 250 requests per hour |
| Bulk organization invitations | Backend API, per Secret Key | 50 requests per hour |
| User/organization/membership updates | Backend API, per resource | 10 requests per 10 seconds, per resource |
Sources: Clerk: System limits
The fix
- Read the Retry-After header on a 429 and wait that many seconds before retrying — do not retry immediately in a tight loop, which just extends the block.
- For the low-ceiling endpoints (sign-in/sign-up creation and verification attempts), add client-side debouncing so a double-click or a fast typo-retry does not count as a second attempt within the same 10-second window.
- Batch operations like invitations respect their own hourly ceiling (100/hour, 25/hour for bulk) — queue and spread large invite lists instead of firing them all at once.
- Use the Backend API's bulk endpoints where one exists instead of looping single-resource calls, since each single call against the same resource counts toward the 10-requests-per-10-seconds update limit.
Sources: Clerk: System limits
How Scribase avoids a comparable wall
Scribase Auth is part of your own Postgres project rather than a shared multi-tenant API with per-application-instance ceilings published up front; sign-in and sign-up calls go to your project's own /auth/v1 endpoint. This page does not claim Scribase has no rate limits at all — only that there is no published 429 ceiling on auth-specific calls the way Clerk documents one today.
Sources: Scribase pricing
Questions people ask
What does Clerk's 429 actually block?
Only the specific endpoint you hit the limit on, for the number of seconds in the Retry-After response header — not your whole application instance.
What is the general Backend API limit?
As documented by Clerk: 1000 requests per 10 seconds on production instances, 100 requests per 10 seconds on development instances.
Why do I hit the limit even though my traffic is low?
The general 1000/10s ceiling is rarely the problem. Several specific endpoints (sign-in/sign-up creation and verification attempts) are capped much lower — 5 or 3 requests per 10 seconds — and a retry loop, a double-submitted form, or an automated test suite hits those far sooner.
Does Clerk's SDK expose the Retry-After value automatically?
Clerk's docs say the header is present on the raw HTTP response, but reports from Clerk's own GitHub issue tracker note the JavaScript SDK does not always surface it in the thrown error object — read the raw response header directly if you need the exact retry delay.