This issue completely stops me from working. Please escalate this to the engineering team as a top-priority bug. Several users are experiencing similar crashes after recent updates.
Hi there…Hello. I have the same problem, where after the last update, Cursor crashes with the above error practically every 15 minutes, making it completely unusable for continuous work. None of the above solutions work. Please treat this issue as a high priority. I would hate to have to abandon Cursor. Thank you.
Need help. I am also facing the same issue on Macbook M2 PRO, sequoia 15.5. It is working fine on Linux Ubuntu though. cursor is keep crashing sometime in 2-3 min after agent starts working which is making CPU usage around 50% on M2. I don’t think it should crash on 50-55% CPU usage proving enough RAM (7-8GB available). It started after an update I did last week which i think i was delaying since Dec 2025.
I am facing it even after the latest updates of Cursor. Will try to rename folder as suggested in one thread. Hope it won’t break any of my settings.
Same here starts after last update to 2.3.39, now we are on 2.3.40 ans still crash. Not a good reputation!
My research. Dev Tools say tons of errors but when the ■■■■■ begins ist logs that:
workbench.desktop.main.js:13688 [composer] Extension host became UNRESPONSIVE undefined
workbench.desktop.main.js:55 ERR [Extension Host] (node:69662) [DEP0040] DeprecationWarning: The `punycode` module is deprecated. Please use a userland alternative instead.
(Use `Cursor Helper (Plugin) --trace-deprecation …` to show where the warning was created)
workbench.desktop.main.js:12674 [Extension Host] (node:69662) [DEP0040] DeprecationWarning: The `punycode` module is deprecated. Please use a userland alternative instead.
(Use `Cursor Helper (Plugin) --trace-deprecation …` to show where the warning was created)
2
workbench.desktop.main.js:13831 [Background Composer] Error updating branch on git status change: Error: No full commit provider registered
at Svs.getCurrentBranch (workbench.desktop.main.js:645:34504)
at pMt.value (workbench.desktop.main.js:13831:5937)
at Ce._deliver (workbench.desktop.main.js:49:2962)
at Ce._deliverQueue (workbench.desktop.main.js:49:3053)
at Ce.fire (workbench.desktop.main.js:49:3390)
at wnu.$gitStatusWasRun (workbench.desktop.main.js:12696:36169)
at JXe._doInvokeHandler (workbench.desktop.main.js:13633:23170)
at JXe._invokeHandler (workbench.desktop.main.js:13633:22912)
at JXe._receiveRequest (workbench.desktop.main.js:13633:21544)
at JXe._receiveOneMessage (workbench.desktop.main.js:13633:20361)
at pMt.value (workbench.desktop.main.js:13633:18388)
at Ce._deliver (workbench.desktop.main.js:49:2962)
at Ce.fire (workbench.desktop.main.js:49:3283)
at jyt.fire (workbench.desktop.main.js:12691:12156)
at MessagePort. (workbench.desktop.main.js:15681:18406)
workbench.desktop.main.js:13688 [transport] Connect error in unary AI connect ConnectError: [unauthenticated] You must be a team member to access this resource
at t (workbench.desktop.main.js:12706:476909)
at async Object.getTeamCommands (workbench.desktop.main.js:566:27065)
at async A9o.loadTeamCommands (workbench.desktop.main.js:15684:3633)
workbench.desktop.main.js:15684 Failed to load team commands: ConnectError: [unauthenticated] You must be a team member to access this resource
at t (workbench.desktop.main.js:12706:476909)
at async Object.getTeamCommands (workbench.desktop.main.js:566:27065)
at async A9o.loadTeamCommands (workbench.desktop.main.js:15684:3633)
This issue is still on going…. Cursor is not usable for at least 10 days now (after an update).
It crashes constantly at nearly every requests (Error 5)…. I’ve perform the various fixes (especially the one about reducing the size of cursor rules) but I’m NOT clearing my context history…
I’ve been building 10 hours a day for 3 months and no problem… Now I can’t use it anymore, like at all….
I tasked Cursor debug agent with finding the most likely cause of the crashes, and this is what it found:
The failure is almost certainly invalid JSON when the Shell tool is called with required_permissions.
What’s going wrong
The Shell tool is invoked with something like:
required_permissions: [‘network’]
or required_permissions: [‘all’]
In valid JSON every string must use double quotes: [“network”], [“all”].
The error mentions things like “ssions”: [‘network’] and ‘all’, 'ne" — i.e. single-quoted strings (‘network’, ‘all’).
A JSON parser that expects strict JSON will then throw Unexpected token ‘’ (the single quote), which matches what you’re seeing.
So the cause of the tool failure is: the payload that runs the Shell command is being built or passed as JSON that uses single quotes (or is truncated), so the JSON parser fails.
That could be due to:
How the agent’s tool call is serialized (e.g. Cursor or the agent runtime turning the tool arguments into a JSON string and accidentally using single quotes or cutting the string off).
How the model formats the tool call (e.g. it outputs [‘all’] instead of [“all”] and the pipeline doesn’t normalize it to valid JSON).
Truncation (e.g. the payload is cut mid-string, so you get ‘all’, 'ne… and the JSON becomes invalid).
So: the JSON error is caused by the Shell tool’s invocation payload not being valid JSON (likely single quotes or truncation). The fix has to be on Cursor’s side (serialization/normalization of tool calls, or better error handling so a bad payload doesn’t crash the app). You can report: “Shell tool fails with JSON parse error when required_permissions is used; payload may be using single quotes or truncated; then Cursor often crashes.”
It seems that it’s another issue with AI hallucinating and doing whatever it wants.
Same issue here. I get the crash even after Cursor does a grep on files, updates audit file, completes, and I hit Keep. Sometimes it just dies half way through. Something really bad is going on. p.s. I have no extensions and did the whole clear cache as well as uninstall/reinstall.
The fix for me was moving my project out of an iCloud-synced folder.
Cursor tends to crash when the workspace is inside Desktop or Documents if those are synced with iCloud.
Solution:
Create a local folder outside iCloud, e.g. ~/dev
Move your project there
Reopen it in Cursor
After moving the project out of iCloud, the crashes completely stopped.
same issue here. Cursor crashes every 2 minutes and I tried the suggested fixes in this thread as much as I could, nothing helped. I thought upgrading to a higher tier might help cuz I wonder if it was my running out of allocated usage causing it but even now I am a pro plus member, it crashes non stop.
I am on MacBook Pro and my project is in local folder (not in an iCLoud-synced folder)
I have the same issue, and because of this, my API requests consume too many tokens, causing my plan to run out quickly. I use ChatGPT Codex, and it reconnects every 2 minutes, causing the program to crash. As a result, my prompt remains incomplete. I have to reopen it and ask again, which means I’m spending 2x the prompts. I’m using a MacBook with an ARM M4 chip, and I haven’t been able to resolve this issue at all. This month, I spent nearly $100 in 10 days on Cursor, and if this continues, I will never use it again.
I’m having the same crashes here. Seems to work fine for a while then gets into a state where it crashes every few minutes. When it reloads I tell it that it crashed and it spends some time (and tokens I bet) figuring out what got done so that it can continue were it left off. In one case when I reviewed its changes it had missed a chunk of work so I think that was due to the crash recovering missing a step. In all I’m very happy with the Cursor features but this constant crashing (and the fact others have been experiencing this for a while) is making me think whether I should switch to something more stable. I sent my logs and screenshots and I am on the Ultra plan.
