Chrome plugin remains disconnected and @Chrome tool is not exposedThis blocks my main use case. I need the Chrome plugin to reliably control signed-in Chrome pages before each task, including reading page content, clicking, filtering, and taking screenshots. Falling back to Windows window-title reading is not sufficient.blocks my main use case”,让
What version of the Codex App are you using (From “About Codex” dialog)?
26.623.31921 Microsoft Store package: OpenAI.Codex_26.623.4041.0_x64
What subscription do you have?
ChatGPT Plus
What platform is your computer?
Windows 11 x64 Microsoft Windows NT 10.0.26200.0
What issue are you seeing?
The Codex Chrome plugin remains disconnected and @Chrome does not expose the dedicated Chrome tab/control interface.
Feedback ID:
no-active-thread-019f0362-7838-7292-84b8-a482828d0593
Chrome extension:
- Codex Chrome Extension v1.1.5 is installed and enabled.
- It is installed in the Chrome Default profile.
- The Chrome extension popup still shows Disconnected.
Native host checks:
- Native host manifest exists.
- Registry key exists:
HKCU\Software\Google\Chrome\NativeMessagingHosts\com.openai.codexextension
- The manifest points to a valid extension-host.exe.
- check-native-host-manifest reports correct: true.
- check-extension-installed reports installed: true and enabled: true.
What happens:
When I run:
@Chrome 读取当前 Chrome 页面标题
Codex does not use the Chrome tab/control interface. It falls back to reading the Windows Chrome window title and says it did not get the dedicated Chrome tab interface.
I already tried:
- Restarted Chrome and Codex.
- Restarted the computer.
- Removed and re-added the Chrome plugin in Codex.
- Reinstalled Codex from Microsoft Store.
- Reinstalled/moved Codex to C drive.
- Started a new Codex thread and tested @Chrome again.
Expected behavior:
@Chrome should connect to the Codex Chrome Extension and read the current Chrome tab/page title directly.
What steps can reproduce the bug?
Feedback ID: 019f0342-9819-7473-ba2c-a4fc426b429a
What is the expected behavior?
_No response_
Additional information
<img width="914" height="505" alt="Image" src="https://github.com/user-attachments/assets/b08328ac-fbd9-4113-93f4-2a78bf4fa666" />
<img width="778" height="450" alt="Image" src="https://github.com/user-attachments/assets/e7aea57b-8430-4149-97a2-be2ac21aad85" />
3 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Additional finding:
After reinstalling Codex and restarting/logging out/in multiple times, the issue still persists.
Local checks show:
[features]
js_repl = false
Because of this, the dedicated Chrome/browser control tools are not exposed to the agent. @Chrome falls back to reading the Windows Chrome window title instead of using the Chrome tab interface.
This does not look like a Chrome extension installation issue anymore. It looks like the Codex app/account/tool exposure layer is disabling or failing to expose node_repl / browser-use / Chrome tools.
I maintain WinBridge Recovery, an independent Windows recovery/diagnostic tool for Codex Desktop local plugin/runtime state.
The combination here is useful: the Chrome extension is installed/enabled, the Native Host manifest and HKCU registration validate, reinstall/restart does not help, and the app later rewrites
config.tomlwithjs_repl = false. That makes it important to separate two very different failure classes instead of repeatedly reinstalling Chrome.WinBridge can verify the Windows local-state side: active
CODEX_HOME, current Store/AppX package generation, bundled marketplace, Browser/Chrome/Computer Use cache versions andlatest,node_repl/CODEX_CLI_PATHconsistency, Native Host manifest/registry/v2 metadata, and the local runtime/app-server paths. If those user-owned layers are stale or version-split, it can reconstruct them from the already installed official package with verification, backup and rollback.However, if Desktop is intentionally rewriting
js_repl = falseafter startup because of a product/entitlement/session provisioning decision, WinBridge should not override that as a security/feature-gate bypass. In that case a clean local-state result is useful evidence that the remaining problem belongs in the upstream tool-exposure layer.So this issue is a good A/B case: diagnose/repair local drift first; if the app still rewrites the feature state and
@Chromeremains unavailable, stop local repair and report that exact behavior upstream. If anyone tests WinBridge on this symptom, outcome reports (worked / partial / did not work) are welcome here: https://github.com/zemeng5208/winbridge-recovery/issues/4