render-monitor

Run a five-step health check: list services for status, list the latest deploy, query error-level logs, fetch CPU and memory metrics, and check p95 HTTP latency. A healthy service shows a live deploy, error rate under 0.1%, and p95 latency under 500ms.

Use the list_logs MCP tool with the service ID, filtering by level such as error, by HTTP status codes like 500 or 503, by search text, or by start and end time. As a fallback, the Render CLI supports render logs with --tail for streaming.

Available metrics include cpu_usage, memory_usage, http_latency at p50/p95/p99 quantiles, http_request_count, bandwidth_usage, instance_count, and active_connections for databases. Each metric has documented healthy, warning, and critical thresholds by plan tier.

Partially. The Render CLI covers service status, deploys, and logs as a fallback, but metrics queries and direct PostgreSQL queries require the Render MCP server configured with an API key at https://mcp.render.com/mcp.

Use query_render_postgres with the postgres ID to run SQL against pg_stat_statements, ordering by mean_exec_time to surface the slowest queries. You can also check active_connections metrics to detect connection pool pressure.

CPU above 85% and memory above 90% are critical; CPU 70-85% and memory 80-90% are warnings. A memory usage to limit ratio above 0.95 signals imminent out-of-memory risk requiring scaling or restart.