-
Notifications
You must be signed in to change notification settings - Fork 6.4k
Description
What feature would you like to see?
Add persistent session recovery to the codex-reply MCP tool by implementing resume_conversation_from_rollout(session_id), similar to the existing app-server implementation.
Current Behavior(AS-IS)
The codex-reply tool currently only calls conversation_manager::get_conversation(conversation_id) and returns a "not found" error if the conversation doesn't exist in memory. This makes the MCP server stateful and fragile.
Desired Behavior(TO-BE)
When a conversation is not found in memory, the tool should attempt to restore it using conversation_manager::resume_conversation_from_rollout(session_id), following the same pattern already implemented in the app-server.
Additional information
Since the app-server already has resume_conversation_from_rollout implemented, this feature would bring the MCP tool to feature parity. It's completely natural for MCP server processes to terminate when the MCP host (like VS Code or Claude Desktop) shuts down - this shouldn't result in lost conversation state. Users expect to resume their work seamlessly when they restart their development environment.