Agents Window: after Remote SSH disconnect, agent dies silently with no reload/reconnect option

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

When the SSH connection to a remote host drops (or the remote extension host loses its Connect transport), agent sessions in the Agents Window fail immediately with:

No Connect transport provider registered.

There is no user-visible notification, no error toast, and no option to restart/reload/reconnect in the Agents Window.

The classic Editor Window has Developer: Reload Window, which recovers the same class of failure. The Agents Window does not expose an equivalent recovery path, so the session just dies and the user only discovers it by inspecting the dead chat.

This happened twice in ~10 minutes on 2026-07-30 with no UI feedback either time.

Steps to Reproduce

  1. Open Cursor Agents Window connected to a Linux host via Remote SSH.
  2. Start an agent chat and leave it running (or idle mid-session).
  3. Cause or wait for the SSH link to drop (sleep, network blip, remote host brief disconnect).
  4. Observe: agent turn ends; chat shows nothing useful / appears stuck or empty.
  5. On the remote host, agent transcript is only:
    {"type":"turn_ended","status":"error","error":"No Connect transport provider registered."}
  6. Cursor Agent Worker logs repeatedly:
    Agent CLI prefetch failed: Error: No Connect transport provider registered.
    Failed to read remoteControlEnabled setting; ... No Connect transport provider registered.
  7. Classic IDE window still offers Reload Window; Agents Window offers no equivalent.

Expected Behavior

After SSH / Connect transport loss in the Agents Window:

  1. Surface a clear error (toast or inline banner): connection/transport lost.
  2. Offer Reconnect / Reload (parity with Editor Window Reload Window).
  3. Optionally offer Retry last turn once transport is healthy again.
  4. Do not leave a silent dead session with no recovery affordance.

Operating System

MacOS

Version Information

IDE:
Version: 3.13.25
Commit: 31e8d61c448c7472e371505838a0fe34083dad50
x64

CLI:
CLI Version 2026.06.04-5fd875e
OS linux (x64)
Subscription Tier Pro

Additional Information

  • Related known error class: No Connect transport provider registered (extension host / cursor-socket Connect RPC). Seen on forum for Editor Window freezes; workaround there is Reload Window — which Agents Window lacks.
  • Related Agents Window + SSH threads (different symptoms, same surface): SSH bugs/parity gaps in Agents Window / Glass.
  • Evidence from remote logs (anysphere.cursor-agent-worker):
    • 06:46–06:56 UTC repeated prefetch / remoteControlEnabled failures with the same Connect transport error.

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Hey, thanks for the detailed report. The transcript and logs really helped.

A couple of quick recovery notes that should unblock you right now:

  • Reload Window does exist in the Agents Window: Cmd+Shift+R or Ctrl+Shift+R, or via the command palette → Developer: Reload Window. It fixes the same class of issues as in the classic Editor Window.
  • Reconnect should be automatic: Glass tries to reconnect Remote SSH when the window regains focus, and the workspace connection status also has an explicit Reconnect.

Your main point is spot on though. When the transport drops in the middle of an agent turn, the session ends with the raw error No Connect transport provider registered. with no toast, banner, or hint to reload or reconnect. That shouldn’t happen, and it’s not something in your setup. We know about this issue and we’re tracking it. I can’t share an ETA yet, but I’ll post updates here.

For now, the workaround is: if you suspect SSH blinked out due to sleep or a network hiccup, do Reload Window or force Reconnect in the connection status before continuing the turn. Let me know if reconnect doesn’t work even manually, that would be a separate signal.

Thank you! Appreciate the quick reply.

Strange, but these workarounds didn’t work:

  1. Close the Agents tab, reopen from the sidebar
  2. Cmd+Shift+R

Quitting Cursor worked.

Thanks for testing and coming back with results. This is exactly the separate signal I was asking for.

The fact that Reload Window Cmd+Shift+R and reopening the Agents tab didn’t restore the session, and only a full Quit helped, is an important detail. It suggests the transport isn’t getting re-registered after the SSH drop inside the window, and the “soft” recovery steps don’t bring it back up. I’ve shared this with the team as extra info on top of what we’re already tracking in this thread. No ETA yet, but I’ll post here when there’s an update.

For now, the workaround is unfortunately what you already found, a full restart of Cursor. If you hit this again, please send the Request ID from that session right top corner of the chat > Copy Request ID, and note whether the workspace connection status showed an explicit Reconnect at the time. That’ll help narrow down where the recovery is getting stuck.