cursor always Reconnecting and Taking longer than… expect…,How should I solve it
And it often breaks my source code encoding
The connection was interrupted repeatedly without saving progress, so the agent stopped to avoid repeating the same work. Retry to continue from the last saved state. Request ID: 06a13685-3c21-4de6-874f-9aaf922592af
Yes, I also encounter this issue. I thought because of I am in China. but if I use USA VPN, it still has the same issue. Now I want to refund 
Connection failed
The connection failed 10 times. Please check your network connection and try again.
Copy Request (daa4ab41-fc1d-4e83-b313-705f5ef3c4d3)
Sure,This is a very terrible thing.I am in china too.
I’ve been having similar issues for the last 24+ hours. Everything is "taking longer than expected ". I have to constantly close and re-open the IDE. I didn’t see any acknowledgement or updates on the Cursor status page (Cursor Status)
Hey everyone, thanks for your patience.
The connection issue that affected networks in China, mainly China Telecom, on the morning of September 23 Beijing time was resolved the same day. On September 24 during business hours, some users in China again saw slowdowns and response dropouts, this has also returned to normal.
If you still see Reconnecting or Taking longer than expected, switch to HTTP/1.1:
- Cursor Settings
Ctrl+Shift+J - Network section
- HTTP Compatibility Mode → HTTP/1.1
- Fully close Cursor and reopen it, then start a new chat
@lihengdao, for refunds, please email [email protected] from the email on your account. A separate team handles that.
@Cpp3DEngine, the file encoding issue is a known issue we’re tracking. Can you share what encoding those files use, like GBK or UTF-8 with BOM. I’ll pass the details to the team.
@francoisTFP, your case looks different from the China network incident. Next time it happens, please send the Request ID (chat menu > Copy Request ID) and your Cursor version, and we’ll check it separately.
I am getting this in Canada. Can work fine on cloud, but cannot get any response in IDE. okay updated cursor, now it works. Why would it stop working though?
Hey @alanbaljeu, glad to hear it started working after the update. This usually happens because of a short network hiccup somewhere between the client and our servers. The requests get through, but the response stream gets cut off, which is why you see Reconnecting and Taking longer than expected. Cloud keeps working because it uses a different route. The update most likely helped because it re-established the connection on a fresh version, not because of a specific fix.
If it happens again, try switching the transport:
- Cursor Settings
Ctrl+Shift+J - Network
- HTTP Compatibility Mode → HTTP/1.1
- Fully close Cursor and reopen it, then start a new chat
If it hangs again, send the Request ID right corner of the chat menu > Copy Request ID and your Cursor version, and I’ll take a closer look.
This issue has been ongoing for weeks and simply will not stop. I experience it primarily with Kimi K3 all the time; I do not know whether it also affects other models.
I simply do not understand how a multimillion-dollar company is unable to resolve issues like this.
The entire error-handling experience is also not user-friendly at all. People are constantly creating posts here in the forum because they see messages such as “Taking longer than expected” or “Reconnecting,” and nobody knows what is happening.
We continuously have to send request IDs because there is zero transparency, and Cursor employees have to investigate the issue themselves based on the request ID. From a developer-experience perspective, that is understandable, but at the very least, proper error handling could be implemented, just as it is in software development.
This means that you do not simply show the end user or the consumer of an application a specific message stating that an error occurred. You should at least provide information about the general category of the problem—for example, whether the provider is currently overloaded or whether there are other issues.
For years, there has been zero transparency regarding these errors, whether internal tools are being called or external providers are being called. As a result, these forum posts keep appearing, nobody really knows what the actual problem is, and we constantly have to wait for Cursor employees to tell us what the issue might potentially be.
Hey, thanks for the detailed feedback. There are some solid points here.
First, I want to separate two symptoms, since they look similar but have different causes:
- Reconnecting or Taking longer than expected is a transport issue. The request reaches us, but the response stream back to your machine gets cut off. Switching to HTTP/1.1 often helps: Cursor Settings
Ctrl+Shift+J> Network > HTTP Compatibility Mode > HTTP/1.1, then fully restart Cursor and start a new chat. - Chat stops on its own with Kimi K3 (what you described in your thread) is a different case where the turn ends quietly without an error banner. This is an issue we’re tracking.
On error handling and transparency, I agree that just saying an error happened without a category (provider overloaded vs our side vs network) is frustrating, especially when you have to dig out the Request ID every time. I passed this feedback to the team as a request for clearer error messages. I won’t promise timelines, but it’s noted on our side.
To look into your specific Kimi K3 case, when it stops again, send the Request ID (chat menu on the right > Copy Request ID) from the Kimi K3 run, plus your Cursor version. Then I can check the exact session, not just the general pattern.
Why does Cursor frequently mess up the encoding of my C++ files, severely disrupting my workflow? Even files I haven’t actually modified or edited sometimes have their encoding corrupted. I use UTF-8 with BOM. Logically, the editor should preserve the source file’s existing encoding rather than changing it while I’m working; this happens at least a dozen times a day.
There is an even more serious issue: Cursor frequently corrupts the encoding of my C++ files, severely disrupting my workflow. Sometimes, the encoding gets corrupted even in files I haven’t actually modified or edited. I use UTF-8 with BOM; logically, the editor should preserve the source file’s existing encoding rather than altering it while I work. This happens at least a dozen times a day.
Hey @Cpp3DEngine, thanks for the details. Got it, UTF-8 with BOM.
Encoding getting corrupted when the agent edits files is a known issue we’re tracking. I can’t share an ETA yet.
One thing I want to clarify, because it affects the root cause. You said the encoding breaks even in files you didn’t edit. Can you help separate these two scenarios?
- Corruption happens only after the agent wrote or changed something in the project (Write / StrReplace)?
- Or the file gets corrupted by itself, for example just by opening or reopening the workspace, with no edits?
If it’s scenario 2, that’s a different mechanism, and it’s important for me to capture it more precisely. Also, please share your Cursor version Help > About and your OS.
And to avoid mixing this with the original topic in this thread (the connection issue), it’d be best to start a separate thread just for the encoding problem. That makes it easier to track and not lose details. Can you create a new post and drop the link here? I’ll pick it up.
Even files that were not edited at all this time have had their encoding corrupted.
Hey, thanks, got it. Encoding breaks even in files the agent didn’t touch. That’s a separate mechanism from agent edits, and it’s important to pin it down more precisely.
I see you’ve already started a separate thread for the encoding issue. Let’s continue there so we don’t mix it with the connection issue from this thread, it’s easier not to lose details. I’ll reply with the actual substance there.
Quick workaround so you don’t lose files right now. Since everything is UTF-8 with BOM on your side, open Ctrl+Shift+P > Preferences: Open User Settings (JSON) and add:
"files.encoding": "utf8bom",
"files.autoGuessEncoding": false
After that, agent edits should preserve the BOM.
To understand the case with files you didn’t edit: when a file breaks, check the chat and see if there was a terminal or shell command right before that which wrote to the file. Also check if the corruption happens just by reopening the project, without a single edit. That’s the key detail because it separates two different paths.
For the ticket, please post in thread 173279 your Cursor version via Help > About and your OS.
About the concerns regarding your code: this is a local file write issue related to the selected encoding. It’s not related to uploading or sending your code anywhere. If you have Privacy Mode enabled, your code isn’t used for training and isn’t stored on our side.