Thanks for the details and the cool “Revert and continue” workaround - it really helps unblock stuck threads.
This is a known issue with the read_file tool - fix is in progress.
For now, recommend:
- Don’t reopen crashed chats (risk of code rollback, see similar case: Cursor failed to read file)
- Keep MCP servers disabled (as you’ve done)
- Work from another workspace or continue via your “Revert and continue”
On context transfer: yes, export chat to Markdown file and import to new chat via @File.
This saves message history and code from old chat. Not full import, but context transfers.
To speed up the fix, could you share:
- Additional Request IDs from failed attempts
- Type/size of file being read, path to it
- Roughly how many messages were in chat before crash
- Did you have code rollbacks when opening crashed chat
Sorry for the delayed response - passed to team. Let me know if the workaround helps.