Bug: Agent stopped retrying repeatedly on Windows standalone desktop app

Describe the Bug

Environment:

  • OS: Windows
  • App: Standalone Cursor Desktop app (NOT VS Code Cursor extension)
  • Request ID: b23a7a9b‑abf0‑4bde‑a020‑ec5bcf305227
  • Cursor Version: 3.17.19 (user setup)

Problem description:
Every time I run Agent tasks to read or modify project files, it pops up “Agent stopped retrying. The connection was interrupted repeatedly without saving progress”.
This error reproduces even with brand‑new windows and brand‑new agent conversations.

Critical observation:
Using the same account, same network, same project, the VS Code Cursor extension works perfectly. No connection interruption error in VS Code extension.

Troubleshooting I have performed:

  1. Fully kill all Cursor background processes via task manager and restart
  2. Set HTTP Compatibility Mode to HTTP/1.1
  3. Clear workspaceStorage cache folder
  4. Add .cursorignore file to reduce indexing overhead
  5. Disable hardware‑acceleration

The bug still persists only on standalone desktop client.

Expected behavior:
Agent should run stably without websocket connection interrupt, consistent with VS Code extension behavior.

Steps to Reproduce

  1. Open Standalone Cursor Desktop on Windows
  2. Open any local project folder
  3. Start Agent task to read or modify project source files
  4. Agent quickly fails and shows error: “Agent stopped retrying”
  5. The issue persists even after opening brand‑new window and new agent conversation.

Important contrast: Same account, same network, same project works perfectly in VS Code Cursor extension.

Operating System

Windows 10/11

Version Information

IDE: Standalone Cursor Desktop
Version: 【你的版本号】
Release Track: stable
OS: Windows 11

Does this stop you from using Cursor

Yes - Cursor is unusable

Hey, thanks for the detailed report with the Request ID and troubleshooting. It makes it clear what’s happening.

The “Agent stopped retrying” message shows up when the agent’s connection to our servers drops before it can save progress. So it stops instead of retrying the same work again. The Retry button continues from the last saved state. From your Request ID, the request was streaming normally and running tools, and then the client connection went quiet. That means the drop is somewhere on the network path between your machine and our servers, not on our side.

Why the standalone app and the VS Code extension can behave differently on the same network: they’re two separate apps with their own settings, so proxy and HTTP mode can differ. On an unstable network path, one might fail while the other keeps working.

What to check:

  1. Compare network settings in both apps, especially proxy and HTTP Compatibility Mode. You have HTTP/1.1 in the standalone app. Check what the extension uses and try setting them the same.
  2. Are you behind a proxy or VPN? If yes, try a different network for a few minutes, like a mobile hotspot, in the standalone app. This will show if the drop is specific to your current network path.
  3. Run Cursor Settings > Network > Run Diagnostics in the standalone app and send the results.

Let me know what the diagnostics show and whether it behaves differently on another network. That will tell us where to look next.

Same toast on Windows standalone. Request ID: b0c22e66-8fa3-4313-b815-92660d7109fa. The agent stopped after connection interrupts during a ~29-file Review batch.

Following the staff note here: open Review first (diffs may already be on disk), then Resume. Next step if it recurs is Settings → Network → Run Diagnostics (and HTTP Compatibility Mode / VPN path).

Hey @villepispa, thanks for the Request ID.

I checked it on our side. The request was streaming and running tool calls normally, then the connection from your machine dropped mid-stream, twice in a row with no progress saved, so the agent stopped. You did everything right: first open Review, then Resume to continue from the last saved state, and we can see in your session that it continued after that with no loss.

The drops are happening somewhere on the network path between your machine and our servers. If it happens again:

  • Run Cursor Settings > Network > Run Diagnostics and paste the results here.
  • Let me know if you’re behind a proxy or using a VPN.
  • Try switching to another network for a couple minutes, like a mobile hotspot. This helps narrow down where the connection is breaking.

Let me know what the diagnostics show.