You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Codex app cross-thread messaging still uses removed dynamic tools #44871
What version of the Codex App are you using (From “About Codex” dialog)?
Version 26.903.71938
What subscription do you have?
pro
What platform is your computer?
No response
What issue are you seeing?
Feedback ID: 019fc9d5-bdc0-7480-97d5-8ba648e5071f
Environment:
Codex desktop app on Windows
Remote Ubuntu project connected through SSH
Codex CLI: 0.154.0
App server: 0.154.0
list_threads and read_thread now work, and previously missing task histories display correctly.
However, send_message_to_thread still fails for every task with:
“This app tool is no longer available through dynamic tools. Use the codex_app MCP server. If that server is unavailable on this host, task delegation is unavailable.”
Expected:
A Coordinator task can send prompts to other existing Codex tasks in the same remote project.
Actual:
The tool appears to route through the removed dynamic-tools implementation instead of the codex_app MCP server.
The issue persists after upgrading the remote CLI/app-server to 0.154.0 and restarting/reconnecting the desktop app. It affects multiple tasks and is not specific to one conversation.
I’ve consolidated my reproduction, clean-runtime test results, and desktop routing findings in the existing headless-host report: diagnostic comment on #42973.
Additional reproduction on 2026-09-23, reported with the affected user's permission.
This also occurs in a Codex desktop task using a local Windows workspace (not a Remote SSH workspace in this reproduction).
Observed:
The model-visible dynamic tool catalog still advertises codex_app__send_message_to_thread and codex_app__wait_threads with their argument schemas.
An explicitly authorized send to an existing task returns the rejection below. Read-only wait_threads with timeoutMs: 0 for the same existing task returns the same rejection.
Repeated checks in subsequent turns reproduce it.
No callable codex_app MCP replacement for these thread operations is exposed in the available tool catalog. Other tools, including local file reads and the GitHub connector used to submit this report, remain usable.
Exact rejection:
This app tool is no longer available through dynamic tools. Use the codex_app MCP server. If that server is unavailable on this host, task delegation is unavailable.
Reproduction:
In an existing Windows desktop task, ask Codex to inspect or send an authorized follow-up to another existing task.
Discover the advertised thread tools.
Call codex_app__wait_threads with a valid existing task ID and timeoutMs: 0, or call the authorized codex_app__send_message_to_thread.
Observe the migration rejection, with no callable replacement available.
Expected: provide the replacement before retiring the advertised route, or return an actionable capability diagnostic/recovery procedure. Existing-task status checks and authorized handoffs should remain available.
Impact: the coordinating task cannot inspect the development task or deliver a prepared follow-up; the user must manually relay it.
Limits: no confirmed root cause, triggering version, or successful restart/update remediation is claimed here. The exact About-dialog app/runtime versions were not collected for this report. No task IDs, project names/paths, conversation transcripts, screenshots, business data, credentials, or raw logs are included. Please clarify the supported way to restore the first-party codex_app bridge for affected existing local Windows tasks.
Additional local-Windows reproduction on 2026-09-23. The affected user explicitly requested that this be reported.
In a Codex desktop task, the model-visible catalog advertises codex_app__list_threads and codex_app__send_message_to_thread, but calling list_threads returns:
This app tool is no longer available through dynamic tools. Use the codex_app MCP server. If that server is unavailable on this host, task delegation is unavailable.
The documented replacement is not usable in this session either: an explicit MCP resource-list request to server: "codex_app" fails during startup with:
MCP startup failed: handshaking with MCP server failed: connection closed: initialize response
Thus the catalog advertises legacy tools that reject calls while the instructed MCP route fails initialization. The user reports that app updates/restarts and a computer restart have not restored inter-task communication. No message was delivered to another task. Please provide a supported recovery path for existing desktop tasks and ensure tool discovery reflects the callable surface.
No task IDs, private paths, project data, credentials, transcripts, or raw logs are included.
Activity
Stix01 commentedon Sep 12, 2026
I’ve consolidated my reproduction, clean-runtime test results, and desktop routing findings in the existing headless-host report: diagnostic comment on #42973.
yjy0701 commentedon Sep 23, 2026
Additional reproduction on 2026-09-23, reported with the affected user's permission.
This also occurs in a Codex desktop task using a local Windows workspace (not a Remote SSH workspace in this reproduction).
Observed:
codex_app__send_message_to_threadandcodex_app__wait_threadswith their argument schemas.wait_threadswithtimeoutMs: 0for the same existing task returns the same rejection.codex_appMCP replacement for these thread operations is exposed in the available tool catalog. Other tools, including local file reads and the GitHub connector used to submit this report, remain usable.Exact rejection:
Reproduction:
codex_app__wait_threadswith a valid existing task ID andtimeoutMs: 0, or call the authorizedcodex_app__send_message_to_thread.Expected: provide the replacement before retiring the advertised route, or return an actionable capability diagnostic/recovery procedure. Existing-task status checks and authorized handoffs should remain available.
Impact: the coordinating task cannot inspect the development task or deliver a prepared follow-up; the user must manually relay it.
Limits: no confirmed root cause, triggering version, or successful restart/update remediation is claimed here. The exact About-dialog app/runtime versions were not collected for this report. No task IDs, project names/paths, conversation transcripts, screenshots, business data, credentials, or raw logs are included. Please clarify the supported way to restore the first-party
codex_appbridge for affected existing local Windows tasks.thunderhog commentedon Sep 23, 2026
Additional local-Windows reproduction on 2026-09-23. The affected user explicitly requested that this be reported.
In a Codex desktop task, the model-visible catalog advertises
codex_app__list_threadsandcodex_app__send_message_to_thread, but callinglist_threadsreturns:The documented replacement is not usable in this session either: an explicit MCP resource-list request to
server: "codex_app"fails during startup with:Thus the catalog advertises legacy tools that reject calls while the instructed MCP route fails initialization. The user reports that app updates/restarts and a computer restart have not restored inter-task communication. No message was delivered to another task. Please provide a supported recovery path for existing desktop tasks and ensure tool discovery reflects the callable surface.
No task IDs, private paths, project data, credentials, transcripts, or raw logs are included.