In your prometheus.yml add a 'remote_write:' block with 'url:' pointing to the remote storage write endpoint (e.g. Cortex, Thanos Receive, or a hosted service endpoint)
Add authentication under the remote_write entry using 'basic_auth:', 'bearer_token:', or 'authorization:' with credentials field referencing a file path rather than inline secrets
Tune queue parameters under 'queue_config:' — set 'max_samples_per_send', 'capacity', and 'max_shards' based on your expected throughput to prevent sample drops
Optionally add 'write_relabel_configs:' to filter or rename metric families before sending, reducing remote write volume for high-cardinality series
Reload Prometheus config with a SIGHUP or POST to /-/reload and monitor the 'prometheus_remote_storage_*' metrics to confirm samples are being sent and not dropped
Known gotchas
Remote write adds WAL (write-ahead log) overhead; if the remote endpoint is slow or down, the WAL can grow large and recovery can lag significantly behind real time
Retries are bounded; if a remote endpoint is unavailable for longer than the retry window, samples are permanently lost — monitor 'prometheus_remote_storage_samples_dropped_total'
Some backends require specific Content-Type or custom headers; missing these causes silent 400/415 rejections that only appear in the Prometheus log, not in standard dashboards
Give your agent this knowledge — and 200+ more routes
One MCP install gives any agent live access to the full route map, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp