Hey, this is a known bug. The composer.composerHeaders key in local storage state.vscdb grows without limits, goes over the Electron IPC threshold, and crashes the panel right when it opens. You have a lot of requests through Composer, which makes it grow faster. The fix is on our side, but there’s no ETA yet.
Workaround is to delete only this key, all other settings and login will stay:
sqlite3 ~/Library/Application\ Support/Cursor/User/globalStorage/state.vscdb "DELETE FROM ItemTable WHERE key LIKE 'composer.composerHeaders%';"
Start Cursor.
The chat history sidebar will be empty, but the chats are stored separately and should load back.
If that doesn’t work, you can do a stronger reset. With Cursor closed, delete state.vscdb and state.vscdb.backup from ~/Library/Application Support/Cursor/User/globalStorage/ after making a backup. This resets some local state, but you can log in again and it should fix it for sure. This worked for another user in this thread:
hanks for the workaround, but unfortunately, it’s not a permanent fix. It only worked for about 30 minutes before the exact same crash started happening again.
Since I use Composer heavily, the composer.composerHeaders key in state.vscdb quickly builds up and hits the Electron IPC threshold again. Running the sqlite3 command fixes it temporarily, but it keeps breaking after a short period of intensive work.
Yeah, that’s expected with your usage pattern. The key fills back up to the IPC threshold as you keep using Composer. The proper fix on our side is still in progress, and I can’t share an ETA yet.
Until then, the only workaround is to run the cleanup periodically. You can save it as a shell alias to make it less annoying, for example add this to your ~/.zshrc:
alias cursor-fix='osascript -e "quit app \"Cursor\"" && sleep 2 && cp ~/Library/Application\ Support/Cursor/User/globalStorage/state.vscdb ~/state.vscdb.backup && sqlite3 ~/Library/Application\ Support/Cursor/User/globalStorage/state.vscdb "DELETE FROM ItemTable WHERE key LIKE '\''composer.composerHeaders%'\'';" && open -a Cursor'
Then just run cursor-fix whenever it crashes again.
I’ve bumped the report internally with a note that heavy Composer usage triggers it within about 30 min, which helps with prioritization. I’ll post here once the proper fix lands.