Remote SSH Agent always times out on HPC cluster — ERROR_EXTENSION_HOST_TIMEOUT / provider registration timeout in Cursor 3.16.29

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

When using Cursor Agent inside a Remote SSH workspace connected to the NeSI HPC cluster, the Agent does not start successfully and eventually times out with ERROR_EXTENSION_HOST_TIMEOUT.

The Remote SSH connection itself works normally. I can browse and edit remote files, use the integrated terminal, and run commands without issues. However, when I send any prompt to Cursor Agent, it keeps spinning and eventually fails with:

Agent Execution Timed Out
ERROR_EXTENSION_HOST_TIMEOUT
The agent execution provider did not respond in time.

The local Cursor Agent works normally outside the Remote SSH workspace.

The local renderer logs repeatedly show that the Agent execution provider fails to register:

[agent-exec] Provider registration timed out (attempt 1/9, timeout=30000ms), retrying
[agent-exec] Provider registration timed out (attempt 2/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 3/9, timeout=5000ms), retrying

[agent-exec] Provider registration timed out (attempt 8/9, timeout=5000ms), retrying

The matching cursor-agent-exec extension exists on the remote Cursor Server, but it does not appear to successfully register the execution provider for the Remote SSH workspace.

Steps to Reproduce

Launch Cursor on macOS.
Connect to the NeSI HPC cluster using Remote SSH.
Wait for the Remote SSH workspace to open successfully.
Browse or edit remote files to confirm the remote workspace is working.
Open the integrated terminal and confirm commands work normally.
Open Cursor Agent.
Send any prompt, for example: Hello.
The Agent keeps spinning and eventually fails with:
ERROR_EXTENSION_HOST_TIMEOUT
Agent Execution Timed Out
The agent execution provider did not respond in time.
Disconnect from Remote SSH and open a local workspace.
Send the same Agent prompt locally; the Agent works normally.

Expected Behavior

Cursor Agent should work normally inside a Remote SSH workspace, just as it does in a local workspace.

After the Remote SSH connection and Remote Extension Host have successfully started, cursor-agent-exec should activate and register the Agent execution provider so that Agent requests can be processed normally.

Operating System

MacOS

Version Information

Cursor IDE: 3.16.29
Commit: 6246455961129c32969845e00aa25d87ae926ec0
Architecture: arm64
Local OS: macOS

Remote environment:
NeSI HPC
Rocky Linux 9.8 (Blue Onyx)
Kernel: 5.14.0-687.30.1.el9_8.x86_64
glibc: 2.34
Architecture: linux-x64
CPU cores visible to login node: 248
Home filesystem: WekaFS

Additional Information

This issue appears to be specific to Cursor Agent inside a Remote SSH workspace.

Troubleshooting already performed:

  • Local Cursor Agent works normally.
  • Remote SSH connection succeeds.
  • Remote files can be browsed and edited normally.
  • Remote integrated terminal works normally.
  • Fully quit and restarted Cursor.
  • Restarted the Remote Cursor Server.
  • Tested with extensions disabled locally.
  • Confirmed that the matching cursor-agent-exec extension exists on the remote Cursor Server.
  • Confirmed that the local and remote Cursor Server commits match.
  • No SIGSEGV, SIGABRT, OOM, or non-zero Remote Extension Host exit code was observed.
  • Remote Extension Host exited with code 0.
  • HTTPS connectivity from the NeSI remote host is working normally.

Remote connectivity tests:

https://api2.cursor.sh
HTTP=200
DNS=0.014882s
Connect=0.211685s
TLS=0.601028s
Total=0.799139s

https://marketplace.cursorapi.com
HTTP=200
DNS=0.085829s
Connect=0.095157s
TLS=0.205555s
Total=0.401809s

The most relevant local renderer log is:

[agent-exec] Provider registration timed out (attempt 1/9, timeout=30000ms), retrying
[agent-exec] Provider registration timed out (attempt 2/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 3/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 4/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 5/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 6/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 7/9, timeout=5000ms), retrying
[agent-exec] Provider registration timed out (attempt 8/9, timeout=5000ms), retrying

The local Agent works correctly and successfully registers cursor-agent-exec. The failure only occurs in the Remote SSH workspace.

This suggests that the Agent execution provider is not successfully registering in the Remote SSH extension host. The failure occurs before a model request can be processed.

The remote environment is an HPC cluster with Rocky Linux 9.8, glibc 2.34, 248 visible CPU cores, and the home directory hosted on WekaFS.

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Hey @Eric_Xue, thanks for the thorough report!

This is a known issue we’re tracking where the agent execution provider fails to come up in some Remote SSH environments (HPC login nodes in particular), and I’ve added your report to it.

Two things from the remote side would help us narrow down where it stalls:

  1. Confirm there’s an anysphere.cursor-agent-exec-* folder under ~/.cursor-server/extensions/ on the cluster
  2. In the Remote SSH window, open the Output panel and copy the contents of the “Cursor Agent Exec” channel (if it’s missing from the dropdown, that itself is useful to know).