Return all allowlisted monitoring checks available to the MCP server.

Response:

• checks[].id: stable check id

• checks[].name: display name

• checks[].description: human-readable check description

Return configured Prometheus environments and their base URLs.

Response:

• environments[].key: environment key (for example prod, dev_test, dr)

• environments[].prom_url: Prometheus URL for that environment

List monitored servers detected from recent up{server_name!=""} series.

Inputs:

• environment: explicit environment key (highest priority).

• env_hint: fallback environment hint when environment is not provided.

Behavior:

• Queries last 10 minutes of up{server_name!=""}.

• Returns unique targets by (instance, job) with server_name.

Return process group names from process monitoring metrics.

Source metric:

• namedprocess_namegroup_cpu_seconds_total{job="process_monitoring"}

• label queried: groupname

Fetch active alerts from Prometheus /api/v1/alerts.

Filters (exact-match):

• severity/state/alertname/job/server_name/instance

Run one allowlisted check via Prometheus query_range and return summarized results.

Inputs:

• check_id: required check id from domain.checks.CHECKS.

• hours/minutes/days: relative lookback window.

• start_time_utc_iso/end_time_utc_iso: absolute UTC range (if provided, this is used).

• end_offset_minutes/end_offset_hours/end_offset_days: shift end time to the past.

• step: range-query step (example: 1m, 5m, 15m).

• include_samples: include raw samples in each series summary.

• server_name: label filter for server_name.

• instance: label filter for instance (example: host-or-ip:9100). Use this when targeting one exact exporter endpoint.

• environment/env_hint: environment selector (environment has higher priority).

Filter behavior:

• If both server_name and instance are provided, both filters are applied.

• If only one is provided, only that label is applied.

Run all allowlisted checks in parallel for the same time range and filters.

Inputs are equivalent to run_check, including:

• server_name: filter by label server_name

• instance: filter by label instance (single-target filter)

Note:

• step is fixed to 5m in this tool to control payload size. Any provided step value is ignored.

Run custom PromQL.

Guardrail:

• approved must be True before execution.

Modes:

• instant=True: use /api/v1/query at a single timestamp.

• instant=False: use /api/v1/query_range for a time window.