How do I recover OpenCode remote sign-in without recreating the Montage server?
OpenCode can list a remote server while its Montage authorization is incomplete or rejected. Recover through the app’s documented sign-in controls instead of creating another entry, and validate the account in the same session that needs workspace access. A server in opencode mcp list can still lack usable account authorization. Use /mcps for the installed app’s supported sign-in controls and verify the recovered workspace in the same affected session.
Help me recover OpenCode remote sign-in without recreating the Montage server. Read whether OpenCode lacks the server, cannot initialize HTTP or receives an auth rejection from a tool call. Compare the actual returned workspace with the intended account. No writes.Identify the failing layer
Read whether OpenCode lacks the server, cannot initialize HTTP or receives an auth rejection from a tool call. Only the appropriate layer should drive the recovery intervention. A missing entry and a rejected grant require different fixes, so preserve the observed stage instead of starting another sign-in by default.
Keep correct registration
Inspect mcp.servers.montage in a compatible v2 project and preserve a correct endpoint. A server displayed by opencode mcp list is not necessarily ready for private workspace reads. The existing entry’s remote type and endpoint should remain stable while testing the account intervention, avoiding a second source of configuration change.
Use the app’s remote controls
Open /mcps and follow the supported sign-in workflow. Review the Montage account during consent; the model provider’s login is a separate authorization system. Use /mcps as documented for the installed app rather than adding guessed CLI auth flags that were never established by the official recipe.
Retry status only
Return to the affected session and request signals_status without resuming a broad old agent task. OAuth recovery should not automatically trigger pending workspace writes or research operations. A retry brief should leave old research instructions dormant so restored consent cannot immediately trigger pending watchlist or audience mutations.
Accept response-based recovery
Compare the actual returned workspace with the intended account. If the server remains listed but rejected, report that distinction rather than calling the registration list a successful connection receipt. The recovery note should distinguish listed, signed in and executed, especially when the UI is reassuring but the actual status read remains rejected.
Keep the boundaries clear.
- Use /mcps as documented for the installed app rather than adding guessed CLI auth flags that were never established by the official recipe.
- The recovery note should distinguish listed, signed in and executed, especially when the UI is reassuring but the actual status read remains rejected.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-15. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.