[Feature] Auto-retry with exponential backoff on "Server is temporarily limiting requests" instead of halting the session

Resolved 💬 2 comments Opened Jun 24, 2026 by Yixn Closed Jun 24, 2026

What I'm asking for

When a session hits:

API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited

Claude Code should automatically retry the request with exponential backoff instead of giving up and halting the turn. Right now the session just stops dead and I have to manually go into every running session and tell it to continue.

This is explicitly not a usage-limit error (the message says so) — it's a transient server-side throttle. Transient throttles are the textbook case for client-side retry-with-backoff, so failing hard on the first occurrence is the wrong behavior.

Why it matters

I run multiple Claude Code sessions in parallel. This error reliably fires whenever I switch Claude accounts (and at other times too). Every time it happens I have to babysit and manually re-prompt every session to keep going. That defeats the point of long-running / parallel / autonomous sessions — one transient throttle kills the whole run.

Proposed behavior

  • On this specific "temporarily limiting requests / Rate limited" response, retry automatically with exponential backoff (e.g. 1s, 2s, 4s, 8s … with jitter), up to a sensible cap (configurable max retries / max total wait).
  • Honor a Retry-After header from the server if one is present, instead of the backoff schedule.
  • Show a non-fatal status line (e.g. "Rate limited, retrying in Ns…") rather than ending the turn, so the session resumes on its own.
  • Only surface a hard error to the user after the retry budget is exhausted.
  • Ideally expose the retry budget / backoff cap as a setting for people who want to tune it.

Related issues

This is the fix requested by the pile of existing bug reports for the same error, e.g. #70082, #70183, #68521, #69098, #65731, #64438, #63929. Those report that the error happens; this asks for the specific remediation (auto-retry with backoff) so the session self-recovers instead of stopping.

Environment

  • Triggered most reliably right after switching Claude accounts, on macOS, Opus 4.8.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗