Sentry is wired into NightMarquee through the Next.js SDK on the client and the server, and it is what tells me a scan failed before a user does. Stack traces map back to source on a Vercel deploy without ceremony, and one alert per new issue is the right default for a one-person project: I read every one. I also point Claude Code at Sentry through its MCP server, so an issue can be pulled into the terminal with its trace and fixed without opening a browser. What could be better: working out what a traffic spike will do to the quota is harder than it should be, and the setup wizard leaves more config files in the repo than I would like.
