refine-ui always returns "Anthropic experiencing high load" — instant canned 200, persisting 2+ days (not load)

[MGCapital1]

Summary

21st_magic_component_refiner has returned the canned message "Anthropic experiencing high load. Please try again later." on every call for 2+ days (2026-06-02 → 2026-06-03, 8+ attempts). The response arrives in ~4–5 s with HTTP 200 — far too fast to be a real generation attempt — which suggests the /api/refine-ui Anthropic upstream is hard-failing (model ID / credits / deploy), not actually under load.

Environment

• @21st-dev/magic@0.1.0 via npx -y @21st-dev/magic@latest
• MCP client: Claude Code, Windows 11
• API key: active (see auth differentiation below)

Evidence

Direct calls to the API (key redacted):

Request	Result
POST /api/refine-ui, valid key	200 in ~4–5 s → {"text":"Anthropic experiencing high load. Please try again later."}
Same, no key	401 → Missing API key header (x-api-key)
Same, invalid key	401 → Invalid or inactive API key
logo_search / fetch-ui (inspiration), same key	✅ work normally

So auth passes and only the generation path fails. One probe mid-window briefly returned 400 → {"error":"Request failed with status code 500"} after ~11 s, then it reverted to the instant canned 200 — consistent with a flapping/broken upstream rather than genuine load.

Secondary client bug (masks the failure)

dist/utils/http-client.js never checks response.status, and refine-ui.js returns data.text unconditionally — so server-side errors surface as successful MCP tool results. Agent clients can't distinguish an outage from a real refinement. Suggest checking status and throwing so MCP clients see an error.

Expected

Either a real refinement, or an honest error status/message (quota, upstream failure, etc.).

[angel-carvajal]

Not able to load, is there a fix?