› Sign-in keeps looping — I click Allow, then Claude asks me to connect again
This was a bug on our side, fixed on July 15, 2026. If you still see the loop: remove the Switch connector in Claude, wait one minute, add it again with the exact URL https://www.switchapp.ai/mcp, and sign in once. Claude keeps a short memory of the old broken answer, which is why the one-minute wait matters. If it still loops after that, contact support and tell us the time you tried.
› Error -32001 / "Unauthorized"
Switch returns JSON-RPC -32001 with the message "Unauthorized" when a tool call needs sign-in — just complete the OAuth login. Note: some clients also show -32001 for their own request timeouts, so check the message text: "Unauthorized" means sign in; "timed out" means wait a moment and retry a small call (like list or balance).
› It says the token expired / "requires re-authorization"
A session that started with an old token keeps failing every call. Re-run the connect / sign-in flow to get a fresh token, then fully quit and reopen your client so it loads the new token.
› Connector is stuck — only "Disconnect" shows, no "Reconnect"
Disconnect (or remove the server), add it again with the same URL, sign in again, and restart the client. Key gotcha: a running session will not pick up a token you refreshed in another window — after re-authenticating, fully restart the client so it loads the new token.
› Login keeps failing in a loop (corrupt cached token)
For CLI / bridge setups, clear the cached auth — rm -rf ~/.mcp-auth on macOS/Linux, or delete %USERPROFILE%\.mcp-auth on Windows — then restart and sign in again. Native desktop connectors manage their own tokens internally, so for those just remove and re-add the connector rather than hunting for files.
› I see the tools but every call fails
Listing the tools needs no login (that is why you can see them). The first tool CALL is what triggers the sign-in prompt. Complete the OAuth login and retry.
› Tools never appear after adding the server
Double-check the URL is exactly https://www.switchapp.ai/mcp, that you restarted the client after editing any config file, and that the config file is valid JSON/TOML (a stray comma silently disables every server). Then try a lightweight call like check balance first.
› Request timed out
A timeout means the request reached the server but no answer came back in time — it is not an auth failure. For big jobs (image/video), the job usually keeps running on Switch, so ask to show your last image or list your recent generations instead of re-firing the slow call. For quick calls (list/balance), wait a minute and try one lightweight call first; check that the Switch website itself loads in a browser; and avoid hammering retries — each one just burns time.
› Still stuck after a clean restart, fresh sign-in, and cleared cache
Run the pre-flight check: the URL is exactly https://www.switchapp.ai/mcp, you restarted the client after any config change, and the config file is valid JSON/TOML (one stray comma silently disables every server). If all of that is true and it still fails, the problem is likely server-side — check Switch status and contact support; no client config can fix an unreachable server.
