ThinkFacility

Error messages

Taking longer than expected...

The message

Taking longer than expected...
Cursor 3.21.18 read September 26, 2026Cursoragentstimeout

What it means

Cursor's client waited about 15 seconds without hearing back from the server. In September 2026 builds it often means the request never left your machine at all, so the window is stuck.

What to do

Fully quit Cursor and reopen it (Reload Window isn't enough), then send again in a new chat. If it keeps coming back, set HTTP Compatibility Mode to HTTP/1.1, and on Windows add Cursor.exe as a Defender process exclusion.

You send a message to Cursor's agent, the spinner starts, and after a few seconds the chat shows this and stays there:

Taking longer than expected...

Sometimes it clears after a couple of minutes and the answer arrives. Sometimes it never does. It's one of the most reported problems on the Cursor forum: the main February thread has 23,127 views, a title search finds 54 topics, and a new one opened on September 21, 2026 passed 50 replies in five days.

What Cursor says the message means

A Cursor staffer explained it on March 18, 2026: the message "shows up when the client doesn't get a server response within the timeout around 15 seconds. There can be different causes, from network issues to server delays." So it's a timer. It tells you Cursor is still waiting, and it doesn't say why.

That's why waiting sometimes works. If the server was only slow, the reply lands late and the chat carries on.

The September 2026 version: stuck before sending

In the 3.21 builds the common case is different. On September 17 staff looked up a user's Request ID (the per-message identifier Cursor support uses to find your request) and found it "never left your machine". The turn was "getting stuck locally before anything is sent", which is why Network Diagnostics passed and changing the HTTP setting did nothing for that user.

On September 22 staff said they're seeing it "on both Windows and Mac and are tracking it", and added that "once a window latches into this state every message in it hangs." That's the case where the message shows up on every single prompt, even "hello" in a fresh chat. As of September 26 there's no fixed version announced, and staff said on September 21 they were "still seeing this on the newest builds too".

How to get it moving again

Older advice from the same threads still turns up: resyncing the codebase index (Cursor Settings, Indexing) "fully fixed it for some users" in March, and testing in an empty folder tells you whether one project is the trigger.

When only one model hangs

If Claude hangs while Grok and Composer answer fine, staff said that "points at the request rather than the window". A restart won't help much there. Switch models for now, and report it: right after a hang, run Developer: Capture and Send Debugging Data from the command palette, or copy the Request ID from the chat menu, and post it with the time and your timezone. In March, one user saw it only on GPT 5.4 past about 60% of the context window, and staff's workaround was smaller chunks of work in fresh chats (they said plainly it wasn't a real fix).

Other lines the same feature prints

Match yours against these if the one at the top of the page is not quite it. They come from the same code and mean related things.

  • Taking longer than expected…
  • Taking longer than expected