Connection Error / Stalled Agent requests for hours

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

The Cursor Agent is completely unresponsive on macOS. Any attempt to send a message results in an immediate hang, eventually timing out with a ‘Connection Error: The connection stalled’ toast notification.

This issue has persisted for over 3 hours. I have attempted to resolve it by restarting the application, creating new chat sessions, and switching between different models (Sonnet, Opus, etc.), but none of these actions have worked. I experienced this exact behavior once before where it spontaneously resolved itself, suggesting a potential intermittent connectivity or server-side issue rather than a local configuration problem.

Steps to Reproduce

Open Cursor on macOS.

Open the Agent / Composer pane.

Type any query (using Sonnet 4.5 or any other model) and hit Enter.

The request loads indefinitely (“stalled”).

After a timeout, the “Connection Error” toast appears.

Expected Behavior

The Agent should process the request and stream the response immediately, instead of hanging and timing out.

Screenshots / Screen Recordings

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.4.2
VSCode Version: 1.105.1
Commit: 7d37854462030d31d936c7f7ed9015b870551e90
Date: 2026-01-20T06:54:16.351Z (9 hrs ago)
Build Type: Stable
Release Track: Early Access
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 25.2.0

For AI issues: which model did you use?

All the models ( Opus 4.5, Sonnet 4.5, GPT 5.2 Extra high, ect.. ) on max mode

For AI issues: add Request ID with privacy disabled

Request ID: 7ce085fe-58ab-4dd7-804e-faf0bd0d6fcf
mje: Connection stalled
at xdf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9095:4486)
at Efa.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9095:7579)
at async Pyt.runAgentLoop (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34199:57735)
at async cpc.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34248:7695)
at async cpc.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34248:8436)
at async DTe.submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9168:74456)
at async Object.Ii [as onSubmit] (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:33000:3808)
at async vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:32974:59943

Does this stop you from using Cursor

Yes - Cursor is unusable

1 Like

Hey, thanks for the detailed report. This looks like a network or connectivity issue.

Try this workaround:

  1. Disable HTTP/2. Add this to your settings.json (Cmd+Shift+P > Preferences: Open User Settings (JSON)):

    "cursor.general.disableHttp2": true
    
  2. If you’re using a VPN, try without it or switch to a different network.

  3. Restart Cursor.

Could you also share a screenshot of the results from Cursor Settings > Network > Run Diagnostics? That’ll help us see at which step the connection drops.

A similar issue was recently fixed this way: Frequent Agent Freezing During Tool-Based Responses. In that case, the VPN was the main cause.

Let me know if it helped.

Hi Dean, thanks for the quick response.

Unfortunately, the issue persists even after following your steps:

  1. I added "cursor.general.disableHttp2": true to my settings.json and restarted Cursor, but there was no change.

  2. I do not use a VPN.

  3. I even tried switching to a completely different internet connection (both are very stable), and the behavior is identical.

I have attached the screenshot of the Network Diagnostics as requested. As you can see, all checks are passing/green (“Healthy”), yet the Agent still stalls immediately.

Is there anything else I can check on my end?

Try a few things:

  1. Reduce the context:
    Try starting a new chat with no files at all, just a simple question. If that works, the issue is the context size.

  2. Roll back to a stable release:
    Cursor Settings > Beta > Stable channel > Restart. You’re on Early Access 2.4.2 (released today), so there might be a regression in the new version.

  3. Console logs:
    Help > Toggle Developer Tools > Console. Send a screenshot if you see any red errors right when it gets stuck.

  4. Temporarily disable your firewall or antivirus (if you use one) and try again. They can block websocket streams even if pings still work.

If none of this helps, I’ll pass both request IDs to the team for a deeper backend review, since the “diagnostics are green but it doesn’t work” situation is unusual.

I’ve send a “hello” in a new chat still same problem and got these erros during and after the request, concerning the antivirus I dont have any. I’ve switched to the stable version I’m waiting for the installation proposal and keep you updated

Currently on default version, still same issue with the stalled connection,

actual cursor version:
Version: 2.4.2

VSCode Version: 1.105.1

Commit: 7d37854462030d31d936c7f7ed9015b870551e90

Date: 2026-01-20T06:54:16.351Z (10 hrs ago)

Build Type: Stable

Release Track: Default

Electron: 39.2.7

Chromium: 142.0.7444.235

Node.js: 22.21.1

V8: 14.2.231.21-electron.0

OS: Darwin arm64 25.2.0

Hello, I managed to unblock the situation, but I had to use a brute force workaround.

The Fix: I simply renamed my project folder in the OS (e.g., from MyProject to MyProject_v2) and opened it as a fresh directory in Cursor. The Agent started working immediately.

Since Cursor treats this as a new project, I have lost all my previous chat history and context which is quite frustrating.

This confirms that the root cause was definitely a corrupted local index/cache specific to that folder (matching the Extension host terminated unexpectedly: Code 5 error in my logs).

For the future, is there a specific command or button to forcefully purge/reset the local index for a workspace without having to rename the directory and lose our chat history? The standard Resync Index didn’t seem to touch the corrupted file.

1 Like

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

“The connection stalled. Please try again.”. Only happens sometimes, depending on what I send to the chat (couldn’t identify any patterns, though)

Steps to Reproduce

Request ID: 0de395a4-4002-46c4-8deb-b8f52d1e669c
Connection stalled
LTe: Connection stalled
at $mf (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9096:4764)
at ova.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9096:8170)
at async qyt.runAgentLoop (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34190:57047)
at async Wpc.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34239:7695)
at async Wpc.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:34239:8436)
at async FTe.submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:9170:14575)

Expected Behavior

Extract relevant info from a .PDF into a .md

Operating System

MacOS

Version Information

ersion: 2.4.21
VSCode Version: 1.105.1
Commit: dc8361355d709f306d5159635a677a571b277bc0
Date: 2026-01-22T16:57:59.675Z (5 days ago)
Build Type: Stable
Release Track: Default
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 25.1.0

Additional Information

I tried:

  1. changing the HTTP config, as seen on other topics. Didnt work
  2. changing project folder name, as seen on other topics. Didnt work
  3. Running network diagnostic, came back with everything OK

Does this stop you from using Cursor

Yes - Cursor is unusable

If Diagnostic is failing then changing HTTP Capabilities under cursor settings worked for me.

2 Likes

This worked! hero

1 Like