What the hell kind of setup do you have such that Cursor has access to delete things in your staging environment?
What happened to you guys ? Don’t feel anxious and keep releasing unstable versions just because your competitors have launched new product features. This will only lead to a significant loss of your existing customers. Your actions have already affected our normal workflow.
i couldn’t agree more. there is no way Cursor should have any number other than 0. in the first position of its version. At best, it is operating at a beta version of quality but feels much more like an alpha.
Our shop conducts a “bugs only” sprint every third sprint. Customers have come to appreciate it. Not sure why there is such a lack of discipline at Cursor. It feels like development is scattered across non-cohesive teams, and no one is parenting the “children”.
@andrewh - can we get a progress update please? As best I can tell, nothing’s been fixed that anyone is reporting, and I’ve just reported more bugs which appeared last week, and I notice a fresh batch of extra mistakes in other forums.
Any chance you can escalate to your manager, and get someone to put a quality-assurance test in-between your developers and your releases? You are LITERALLY breaking 2x more things than you’re fixing with every release.
I went to my meetup Xmas party yesterday. In Feb, 100% of us were using Cursor. Today, only about 10% are left - everyone else had enough. We have 954 members ( Coding from Beach | Meetup ). Your engineering team is actively sabotaging your entire product man!
Here are some things we’ve just fixed
- Plan mode flicker bugs fixed
- Agent review should show less frequently
- Bunch of worktree stability/quality improvements, will come back to you with those, mostly that edits should not fail anymore (a bug where it would say
old_str replace failed) - Some inputbox and menu lags with the @-menu
- Checkpoints bug fixes (these will be released sooner in 2.2.31 in the next day or so)
- Windows legacy terminal behavior should be improved
- Clicking on files in the reviews section should open the full file
- The caching/bedrock post above is not a bug. It will reduce net costs for users.
What else should we focus on? Another priority is that you never have to press “try again.”
Firstly - MASSIVE thanks for that update - we 1000% appreciate this!!!
Please fix the MCP server issues ASAP - non-stdio (SSE/streaming HTTP) is establishing massive numbers of simultaneous duplicate connections (you have a serious threading leak in your code), reconnects for no reason every 20s, doesn’t know all CA intermediate TLS certs (e.g. LE R12), struggles with authentication, incorrectly calls the wrong tools on the wrong servers when different servers happen to have the same tool names, and other mistakes.
A “nice to fix” second easy one to fix would be your hooks:
Your spec says hooks can return this:-
// Output
{
"permission": "allow" | "deny" | "ask",
"user_message": "<message shown in client>",
"agent_message": "<message sent to agent>"
}
but that is never reaching the agent.
Still have The Legacy Terminal option doesn't work anymore after update from v1.7 to v2.x
Version: 2.2.36 (system setup)
VSCode Version: 1.105.1
Commit: 55c9bc11e99cedd1fb93fbb7996abf779c583150
Date: 2025-12-18T06:25:21.733Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.22631
The /summarize command is also broken - it clears the prompt, does the summarization, and that’s it.
We are examining MCP bugs now.
That is the intention of the summarize command - anything else that you were expecting?
@andrewh if I can help in any way - I’m happy to! reply or whatsapp +61414505452
this reminds me - Cursor does not seem to be able to accept large input prompts - 200k models refuse prompts through cursor, which work fine when sent other ways. I think you have a bug in your transport system which can’t take big data inputs from us, and then incorrectly says it’s too big for the model, when the real problem is that it was too big for some other non-model-related part of your input pipeline.
My input was 488322 bytes, 110278 tokens, and every model of yours refuses it (including 200k sonnet and 256k others) using Windows 2.2.20+ IDE
Previously, you could write a prompt and send it by invoking this command. The command would be executed, and then the prompt would be processed. If context compression resulted in poor performance, you could roll back to the command execution and context compression, and run the command with the full context.
Curosr should thank the open-source VSCode, such a stable, reliable, and efficient code editor. Despite making so many almost absurd modifications to VSCode, it actually took them two years to reach the point where bugs kept popping up.
I have no idea how to log this, but Cursor is extremely laggy, compared with VSCode, connected from a Windows host to WSL.
I beg the DEV team to encourage their team to use a Windows + WSL setup, actively.
how big are your chats? I notice the “lag” goes off-the-charts when you use piles of memory (which long chats do). What does task-manager say?
My cursor instance using 2gb RAM takes about 30s for my keystrokes to appear. Other ones using only 200k are instant.
@andrewh - is there any update on when Hooks and MCP connections are going to be fixed? The problems are dire, and have been stretching on for months now.
MCP should be fully fixed in 2.3. Hooks additional events are coming, but should also work. What issues are you seeing?
After reading almost all of the conversation threads, the main point of this discussion is a request to please stop frequently releasing unstable versions to the Stable branch. Every update introduces more uncertainty.
Bug issues and the problem of repeatedly pushing unstable releases are two separate things — why are these being conflated?
Can you explain under what circumstances it makes sense to release updates to the stable version twice within a single day? The details of each update are unknown every time.
Please provide an explanation. If you are unable to answer, could you have someone else respond instead?
We do not update the stable version of Cursor twice in one day. We aim to update once per week, but sometimes twice per week. Could you provide further feedback on bugs or parts of the product?
On what goes into each update: We’ve started moving changelog updates earlier with 2.3. We understand that is helpful to know before updating. Changelog · Cursor