›_ Codex Issues
Codex Issues / #31970

Codex Desktop 26.707.31428: Slack and Atlassian codex_apps MCP handshake times out after update

Resolved 💬 2 comments Opened Jul 10, 2026 by socar-brad Closed Jul 10, 2026

Summary

After updating Codex Desktop, every built-in codex_apps connector call for Slack and Atlassian fails during MCP startup with a 30-second handshake timeout.

The same Slack connector had worked earlier in the workflow, including creating a Slack draft. After the desktop update, both Slack and Atlassian started failing consistently. The in-app Browser plugin still works and can open an authenticated Jira page, so the failure appears scoped to the built-in codex_apps MCP path.

Related: #24785, which reports the same handshake failure on an older desktop version with Gmail and GitHub connectors.

Environment

  • macOS: 26.5.2 (build 25F84)
  • Codex Desktop bundle version: 26.707.31428
  • Desktop build: 5059
  • Codex Framework: 150.0.7871.101
  • codex-cli: 0.144.0-alpha.4
  • Auth mode: ChatGPT
  • Affected connectors:
  • Slack
  • Atlassian Rovo (Jira/Confluence)

Reproduction

  1. Update Codex Desktop to 26.707.31428.
  2. Start a desktop thread with previously working and authorized Slack and Atlassian connectors.
  3. Invoke a read-only Slack tool such as mcp__codex_apps__slack_slack_search_channels.
  4. Invoke a read-only Atlassian tool such as mcp__codex_apps__atlassian_rovo_getjiraissue.

Actual behavior

Both calls fail with:

tool call error: failed to get client

Caused by:
    MCP startup failed: timed out handshaking with MCP server after 29.999999792s

This reproduces across multiple retries and affects both connectors at the same time.

Expected behavior

Codex Desktop should initialize the built-in codex_apps MCP server and invoke the already-authorized Slack and Atlassian connector tools.

Additional observations

  • The failure began immediately after the desktop update.
  • Slack and Atlassian fail with the identical MCP startup error.
  • The in-app Browser plugin remains functional and can access an authenticated Jira page.
  • Local shell tools remain functional.
  • Direct calls to https://chatgpt.com/backend-api/wham/apps were not attempted in this environment, so this report does not claim direct backend reachability.

View original on GitHub ↗

2 Comments

socar-brad · 1 month ago

Update: the issue recovered intermittently in the same Codex Desktop session.

At 2026-07-10 10:39 KST, both read-only calls started succeeding again:

  • Slack channel search
  • Atlassian Jira issue fetch

No connector reauthorization, app reinstall, or additional desktop update was performed between the failing and successful calls. Both connectors recovered together. This makes a transient codex_apps MCP startup/reconnection problem more likely than permanently invalid connector credentials.

The failure had reproduced across several retries before recovery.

socar-brad · 1 month ago

Closing this report because Slack and Atlassian connector calls are currently working again in the same desktop session.

The recovery occurred without reauthorization, reinstalling the app, or installing another update. The incident therefore appears intermittent rather than a persistent credential failure. If the 30-second codex_apps MCP handshake timeout recurs, I will reopen this issue with timestamps and additional logs.