No Agents working and No source control providers registered

Support Bug Reports
,
52

I have the same problem.

53

Hey @gregfunky, this is a really valuable report, thanks for testing. The combo of uncommitted changes plus those changed files being open in the viewer or editor on startup is a strong signal, and it independently matches what @Quang_Hoang_Dinh described in post #28. The fact that closing the open changed files plus restarting brought all windows back also fits the hypothesis that exthost is hanging while restoring the git diff or file on startup.

While the team is looking into it, the most reliable workaround for everyone in the thread for this trigger:

  1. Before closing Cursor with session restore enabled, close tabs with changed git files and or do a commit or git stash in the pending changes view.
  2. If the window is already broken, close the open changed files, then Cmd/Ctrl+Shift+P ā†’ Reload Window.
  3. If that doesnā€™t help, delete the per-workspace state.vscdb only for the affected project not the whole workspaceStorage, as @ivnmood wrote in post #23. This resets only the local UI state for that workspace, it does not touch the repo or settings.

@Malay_Kumar, I get it, 3 to 4 times a day really blocks your work. Rolling back to an older version usually isnā€™t needed. The workaround above closing changed files before restart, or deleting state.vscdb just for the affected project clears the block without a downgrade. If you want to temporarily avoid auto updates, open Cmd/Ctrl+,, search for update, and set updates to manual. I canā€™t share an ETA for a fix yet, but the issue is being actively tracked, and the logs plus findings in this thread directly help.

@sun67, try the same workaround and let us know if it helped. If you can share more details, what OS and Cursor version are you on, and do you hit this specifically when files with git changes are open on startup.

If the bug happens again, logs from the broken window are especially useful main.log, renderer.log, and the exthost/ folder if it exists from ~/Library/Application Support/Cursor/logs/<latest>/window<N>/. Every case like this helps narrow down the cause.

54

  • I have the Codex standalone application (not the extension). But regardless, I donā€™t think the extension is the issue because Iā€™m pretty sure Iā€™ve had this error even why trying to launch Cursor with extensions disabled - not something Iā€™ve done in a while since I now default to launching cursor with my script instead of messing around with other ways.

  • Logs attached - I just did the latest update which relaunched Cursor and the error is still there. So Version: 3.6.31 (Mac OS Latest)

20260602T103628.zip (129.4 KB)

PS - Not sure why this topic is marked as solved, it definitely isnā€™t

55

Itā€™s happened again on Cursor version 3.6.31, Windows build is Windows 10 Home 22H2. After pressing Ctrl-Shift-P I couldnā€™t find an option ā€œDeveloper: Export Logsā€, but I did find ā€œDeveloper: Capture and Send Debugging Dataā€. This creates a zip file ā€œdebugging-dataā€ with those files in - I can send the main.log, renderer.log and exthost/ from the most recent of these?

EDIT: I think Iā€™ve found the problem, there were two Anysphere extensions that always tried to load and failed - one was a WSL one and I think maybe a Python one. Instead of clicking ā€œIgnoreā€ on these I clicked ā€œUninstall and restartā€ and it seems to be working much better now.

EDIT 2: The problem is back again, deleting state.vscdb in the relevant folder fixed it.

56

@hekima, thanks for the archive. A stable repro on every launch is really valuable, intermittent cases are much harder.

The logs confirm the exact signature weā€™re tracking:

  • Only window1 is broken (workspaceId 55a8e9801fbf7304ff0f693814255bd9). The exthost/ folder is completely missing. Four Extension Host processes spawn (the renderer logs Started local extension host with pid ...), but none of them ever reaches the logger. After 30s we hit Timed out waiting for git context provider, then a cascade of No Connect transport provider registered. The process is alive, but bootstrap never finishes. Thatā€™s what leads to No source control providers registered. vscode.git runs in exthost, and exthost never came up.
  • window2 works. Also, the DI error $di$dependencies in the renderer shows up in both windows, so itā€™s noise, not the cause. You were right about that.
  • We can also rule out Codex as an extension. The issue reproduces without it.

One new lead: sharedprocess.log has Error getting extensions control manifest Timeout. window2 survived it, so it might not be the root cause, but Iā€™m looking into it.

While we investigate, thereā€™s an easier workaround than your rm -rf workspaceStorage script. That script resets local UI state for all projects on every launch like tabs, terminal history, layout.

  1. If a specific window is stuck, close only that window, not the whole Cursor app, then reopen it. You usually donā€™t need to delete workspaceStorage for this.
  2. If youā€™re opening 2+ workspaces, try opening them one by one. Let each one fully start up (auth ready, Source Control shows up) before opening the next.

Note: for some users this doesnā€™t always help. If it doesnā€™t work for you, tell me, that signal is useful too.

Since you can repro on every start, you can capture the cleanest diagnostics. When the window is in the broken state (before restarting or running the script), if you have a minute:

  1. The full contents of ~/Library/Application Support/Cursor/logs/<latest-timestamp>/ from that launch. It should include Cursor Helper (Plugin).log from the exthost processes. It may show exactly what itā€™s stuck on during startup. I didnā€™t see those in the archive you sent, there was only the Developer Logs Folder.
  2. A sample from the stuck exthost pid. In Terminal run sample <pid> 10 -file ~/exthost-sample.txt using the pid from renderer.log like 17428. Or use Activity Monitor, select the process, then Sample Process. This will show where the thread is parked.
  3. If available, crash reports from Console.app for the same time window, or a sysdiagnose.

I canā€™t give an ETA for a fix yet, but your stable repro seriously increases our chances of reproducing it locally. Send whatever you can.

59

Hey everyone, quick update on this thread. The fix for the No source control providers registered issue and the extension host init crash when opening a window is included in Cursor 3.8.

Once you update to 3.8, please try again and keep an eye on it, especially with multiple windows and workspaces where this happened more often.

If the issue still reproduces on 3.8, please donā€™t restart Cursor right away. Instead run Cmd+Shift+P then Developer: Export Logs and attach the Window, Main, and Extension Host sections here, plus the crash report from Console.app if there is one. Then we can look at your case separately on the new version.