Agent Execution Timed Out /ERROR_EXTENSION_HOST_TIMEOUT — recurring on macOS, already on latest version

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Cursor version: 3.17.8 (stable channel, already confirmed up to date via Check for Updates)
OS: macOS, Apple Silicon (arm64)
Setup: Working entirely locally (not Remote-SSH — confirmed status bar shows “This Mac”)

Getting recurring “Agent Execution Timed Out” errors (ERROR_EXTENSION_HOST_TIMEOUT). Console shows the extension host repeatedly terminating unexpectedly (Code: 6) while these extensions were running: anysphere.cursor-agent-exec, anysphere.cursor-checkout, anysphere.cursor-retrieval, vscode.git, vscode.git-base. Provider registration for agent-exec times out and retries up to 9 times before failing.

Sample Request ID: 0934d09c-9763-4c47-8259-8c480a095f08

Already tried: force quitting and relaunching Cursor, confirming no update is available. Issue persists.

Steps to Reproduce

Issue appears to occur intermittently, not from a specific reproducible action:

Open Cursor and start a new conversation with the Agent/Composer (chat panel)
Send a message that requires the agent to execute a tool call
Agent returns “No first token received” warnings, followed by repeated “[agent-exec] Provider registration timed out” retries (up to 9 attempts)
Request ultimately fails with “Agent Execution Timed Out” (ERROR_EXTENSION_HOST_TIMEOUT)
Console confirms the extension host process terminated unexpectedly (Code: 6) while anysphere.cursor-agent-exec was running

Not consistently reproducible on demand — recurs periodically during normal use rather than after one specific trigger action.

Expected Behavior

The Agent should execute the request normally and return a response without the extension host crashing or timing out.

Operating System

MacOS

Version Information

Version: 3.17.8 (Universal)
VS Code Extension API: 1.128.0
Commit: 2fdd31c9f33f7fbe501f2d57772dc5bf64b63620
Date: 2026-08-20T02:18:12.724Z
Layout: Agent Window
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.291
OS: Darwin arm64 25.5.0

For AI issues: which model did you use?

auto

For AI issues: add Request ID with privacy disabled

0934d09c-9763-4c47-8259-8c480a095f08

Additional Information

From the console logs already captured:

The specific crash signal: Code: 6, Signal: unknown
It’s not a one-off — you saw it recur multiple times in a single session, with the same [agent-exec] Provider registration timed out pattern retrying up to attempt 9/9 each time before failing
Related errors appearing alongside it: PluginsProviderService Timed out waiting for plugins provider after 30000ms, BrowserSnapshotWatchdog Failed to trigger browser snapshot backfill, and CodebaseTelemetry Trigger snapshot command failed

Worth mentioning if true:

Whether it happens on every agent request now, or only some of the time
Whether it’s tied to a particular kind of request (e.g., ones that involve running a terminal command, editing a large file, or right after your Mac wakes from sleep) — if you’ve noticed any pattern
How long you’d had Cursor open / how many agent messages you’d sent before it first happened that session

Attach if the form allows file uploads:

One or two of the console screenshots you shared with me earlier — the ones showing the “Extension Host” section and the “[composer] Error in AI response” block are the most diagnostic, since they show the actual stack trace

Does this stop you from using Cursor

Yes - Cursor is unusable

This is happening to me across all of my Repos. I submitted a request today for a full refund for my Ultra account after I used it for maybe 4 hours, and then Cursor bricked. It has been fun, but this is ridiculous. I am Windors 11

Request ID: f01f13d1-610c-4a7e-86dc-c1df3aa81c08

Request ID: b337aca2-56fe-4708-900b-eeb66535fc32

2026-08-21 18:11:23.632 [info] [RendererPing] enabled (intervalMs=5000, blockThresholdMs=15000, stacks=true) 2026-08-21 18:11:23.733 [info] updateURL https://api2.cursor.sh/updates/api/update/win32-x64-user/cursor/3.17.8/562a08ce36ec5cda1c7703e1ecb708bd213ce4be16cea606612ef86c89e87aba/stable 2026-08-21 18:11:23.735 [info] update#setState idle 2026-08-21 18:11:34.480 [info] [LocalAgentStorage] Scanned 9 recent databases (skipped 0 old), found 11 agent headers, deduped to 11 2026-08-21 18:11:52.392 [info] [PowerMainService] Started wakelock id=0 owner=window:1 reason=“agent-loop” activeCount=1 2026-08-21 18:11:53.652 [info] [PowerMainService] Disabled background throttling id=1 owner=window:1 reason=“agent-loop” activeCount=1 2026-08-21 18:11:54.892 [info] update#setState checking for updates 2026-08-21 18:12:00.088 [info] update#setState idle 2026-08-21 18:12:05.403 [info] [CursorProclistService] Config enabled feature (subsample every 10s; memoryPressureMonitorEnabled=true) 2026-08-21 18:13:04.694 [info] [PowerMainService] Stopping wakelock id=0 owner=window:1 reason=“agent-loop” heldForMs=72301 2026-08-21 18:13:05.230 [info] [PowerMainService] Stopped background throttling disable id=1 owner=window:1 reason=“agent-loop” heldForMs=71578

Hey there.
I checked both request IDs and neither reached our backend, so this is the extension host on your machine not starting in time, not your account.

To get unblocked now: hit Reload Window on the error (or Ctrl+Shift+P → Developer: Reload Window); fully quit and relaunch if it returns. The usual Windows fix is adding an antivirus/security exclusion for %LOCALAPPDATA%\Programs\cursor\ and %APPDATA%\Cursor\, since AV scanning at launch can stall the extension host.

Since you’re on Windows (this thread’s on macOS), could you open a new thread with your Extension Host logs (View > Output → Extension Host)? That lets us pin your specific cause. The refund’s handled over email, so no need to sort that here.