Cloudflare ships the metrics. Nobody pages you.
Workers cron and Workflows monitoring with the dead-man switch Cloudflare does not have. Read-only token, connected in minutes.
Platforms watched: n8n and Cloudflare Workers/Workflows
Four detections, every five minutes
Cloudflare gives you Workers analytics, Workflows metrics, and logs - and no native failure alerting, no dead-man switch for schedules. A cron trigger that silently stops firing looks exactly like a quiet afternoon. Midwatch watches from outside and pages you when the rhythm breaks.
Silence from the declared schedule
Your cron expression declares the rhythm - Midwatch computes the expected interval from it and pages when a schedule stops firing. No history needed, no learning period: the dead-man switch works from the first cycle.
Hard failures
Exceptions, exceeded resources, and internal errors on Workers invocations and Workflow runs, pulled from the analytics Cloudflare already collects - turned into alerts Cloudflare does not send.
Reachability
Token revoked, API timing out, account moved - one deduplicated critical alert on the connection itself, with per-script checks suppressed while it is down so you get signal, not noise.
Deploy detection
Every script and Workflow modification timestamp is compared against the baseline you blessed. A new deploy raises an alert with the timeline - acknowledge it or investigate, Midwatch never decides for you.
Read-only, enforced by the platform
You connect with a Cloudflare API token scoped to exactly two permissions: Account Analytics: Read and Workers Scripts: Read. Cloudflare itself enforces those scopes - Midwatch cannot write to your account even in principle. Nothing is installed in your Workers, no wrapper around your code, no agent in your account.
Honest limitations
Cron script runs are counted per fire, exactly - Midwatch reads the raw scheduled-invocation events, not a sampled aggregate. Only Cloudflare Workflow outcome counts are subject to adaptive sampling, and only at very high volume. Silence detection is never sampled: it works off your declared cron schedule. Analytics lag a few minutes behind real time, so detection latency is the poll cycle plus that lag. Retention is 31 days, which bounds backfill - the declared-schedule dead-man switch needs no history at all. If a limitation matters to your fleet, ask us before you buy.
Put a watch on your Workers
Same trial, same plans, same console as n8n monitoring - a connected Cloudflare account counts as one instance.