Sentry
Production errors → features. Read-only token, aggregate data, zero SDK changes.
How it works
We read each Sentry event's stack-trace filenames plus the release commit SHA, then join them to the scan that was active at that commit through the feature's path index. Counts and event IDs only — never request bodies, never PII. Connect once via OAuth or an API key on the Integrations page; no code changes.
Evidence it adds
get_feature_errors— production errors mapped to a feature (MCP tool)- Errors per feature over 24h and 14d
- Regression flag — fires when the 7-day error rate is ≥ 2× the prior 7 days
- Kept as its own production-error signal; it is never blended into another metric