Agent tool calls (Shell, Read, Glob) hang indefinitely on Remote-SSH — sporadic for ~1 week

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Environment: Cursor desktop on macOS/Windows laptop, connecting via Remote-SSH to multiple Ubuntu Linux hosts (3 different machines). Latest Cursor version as of 2026-05-01.

Symptoms: Every agent tool call (Shell, Read, Glob) hangs for 5–30+ minutes and never completes. The integrated terminal works perfectly on the same hosts. The issue reproduces on all three remote hosts (different hardware, different filesystems). It started ~Apr 24 and is sporadic — sometimes tools respond in <100ms, then hang indefinitely minutes later.

What I’ve ruled out (with an agent’s help over 6 hours):

Filesystem latency: ZFS home (local), NFS /maps — both respond in <5ms from the terminal
Shell init: standard .bashrc, no blocking commands, conda init returns in 3ms
Duplicate cursor-server processes: killed all stale servers, confirmed single stack, still hangs
Server build mismatch: tried with only one build present, still hangs
Host load: tried on lightly-loaded and heavily-loaded hosts, same behaviour

Workaround: None found. User must paste terminal output manually into the chat for the agent to process.

Steps to Reproduce

Open Cursor desktop, connect via Remote-SSH to a Linux host
Open a workspace on a network-mounted filesystem (NFS)
Start a new agent chat
Ask the agent to run any tool — e.g. “read /etc/hostname” or “run echo hello”
Tool call hangs indefinitely. Cancelling and retrying sometimes works briefly, then hangs again.

Expected Behavior

Agent tool calls (Shell, Read, Glob) should complete within seconds, matching the responsiveness of the integrated terminal on the same host.

Operating System

Windows 10/11

Version Information

Version: 3.2.16 (user setup)
VSCode Version: 1.105.1
Commit: 3e548838cf824b70851dd3ef27d0c6aae371b3f0
Date: 2026-04-28T21:07:47.682Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200

For AI issues: which model did you use?

Opus 4.6, 4.7

Additional Information

Remote server logs available at ~/.cursor-server/data/logs/ on each host. Issue reproduces on 3 independent Ubuntu Linux hosts (different hardware, ZFS+NFS and pure NFS filesystems). The integrated terminal on the same hosts works perfectly — only agent tool dispatch is affected. Chat transcript ID for this debugging session: b22a8c25-867f-4f8b-a7b4-44faeb069a91.

Does this stop you from using Cursor

Yes - Cursor is unusable

Has been happening to me over the last ~dozen Cursor versions as well. Symptoms match the original poster. Client and Remote SSH host are both Ubuntu Linux, using Agent view. It seems worse when I have more agents running.

When the Shell tool fails, it appears to return immediately without actually executing the command, even for trivial commands like date, echo, or redirecting output to a file.

Example failed command:

ls data/cache/btc_v3/single/ > /tmp/check_files.txt 2>&1; wc -l /tmp/check_files.txt

After the tool call, /tmp/check_files.txt did not exist, which suggests the shell command body never ran at all. The same happened with a later simple test that should have written /tmp/shell_check.txt.

Environment details from a working Shell tool call on the same remote:

SHELL=/bin/bash

BASH=/bin/bash

BASH_VERSION=5.2.21(1)-release

login_shell=no

interactive=no

BASH_ENV=

PWD=/root/poly

The Shell tool process is a non-login, non-interactive /bin/bash launched under the Cursor extension host. The visible integrated terminal is separate and uses bash with Cursor shell integration.

Remote host is running Linux 6.8.0-100-generic

Request ID:

df6f6baa-5952-4eb3-90ff-f0bbebd37569