Skip to content

fix: isolate MCP refresh token invalidation - #247

Closed
massimoalbarello wants to merge 1 commit into
mainfrom
codex/isolate-mcp-refresh
Closed

massimoalbarello wants to merge 1 commit into
mainfrom
codex/isolate-mcp-refresh

Conversation

@massimoalbarello

Copy link
Copy Markdown
Owner

Why

#236 gave each agent instance its own authorization, but refresh-token replay detection in the OAuth provider still invalidates every login for the same user and OAuth client. Because Codex instances share a client ID, one instance retrying a stale token can disconnect the others.

How

Pin and patch the provider so replay cleanup is scoped to the originating connection and authorization-code grant, while retaining the user and client boundaries. Preserve token rotation and the 30-second retry window. Extend the real OAuth/SQLite tests for shared CIMD and dynamically registered clients to prove that the affected token family is revoked while another instance keeps refreshing and accessing MCP.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant