Codex Desktop messages remain invisible until restart while MCP approval dialogs still appear
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
What platform are you using?
Codex Desktop on Windows. The affected task used a remote HTTP MCP server with MCP Elicitation. The thread had recently transitioned out of realtime voice mode; it is not yet known whether that contributed.
What issue are you seeing?
New assistant status and result messages stopped appearing in the Codex Desktop timeline, while MCP Elicitation approval dialogs continued to appear and remained functional.
The user could therefore approve privileged MCP commands but could not see the surrounding assistant messages that explained the analysis, proposed scope, and expected effects. After completely restarting Codex Desktop, the previously missing messages appeared.
This is not message deletion and not an MCP confirmation bypass: the approval dialogs were visible and explicitly accepted. The defect appears to be delayed UI synchronization or rendering of assistant messages.
Steps to reproduce
- Open a long-running Codex Desktop task on Windows.
- Use an HTTP MCP server whose tool requests confirmation through MCP Elicitation.
- Let the assistant emit status messages before and after MCP calls.
- Observe that the Elicitation dialogs can appear even though new assistant messages are absent from the timeline.
- Continue the task and approve the requested MCP commands.
- Completely restart Codex Desktop.
- Observe that the previously missing assistant messages now appear.
The issue was observed during a maintenance workflow. Privileged commands were confirmed in Elicitation, but the explanatory messages and visible audit context were unavailable until restart.
Expected behavior
- Assistant messages, MCP calls, approval dialogs, and results should remain synchronized in the visible timeline.
- An approval dialog should never become actionable before all preceding user-facing context has rendered.
- If the timeline loses synchronization, privileged approvals should be blocked or display a warning and enough self-contained context.
- The UI should provide an explicit reconnect or refresh recovery path without requiring an application restart.
Actual behavior
- MCP approval dialogs appeared.
- Assistant messages did not appear.
- Restarting Codex Desktop restored the missing messages.
Safety impact
This can cause a user to approve a privileged operation without seeing the assistant's preceding explanation or target summary, even though the MCP confirmation mechanism itself is working.
Related issue
This resembles the broader Desktop history and rendering family reported in #17354, but differs because newly generated messages existed and appeared after restart, while MCP approval dialogs continued updating in real time.
Local paths, filenames, task identifiers, command contents, and other incident-specific details have intentionally been omitted.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
No files, tests, or code entry points are named. Reproduce on Windows with a long-running task using an HTTP MCP server and MCP Elicitation, then trace how timeline messages and approval dialogs are synchronized. Done means messages remain visible with approvals, or the UI provides the described warning or recovery path without requiring a restart.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- desktop
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 42/100