| name | metrics |
| description | Search and query Grafana metrics (metrics.qlikcloud.io) to investigate a ticket - find the relevant dashboard, read its PromQL, run and adjust queries, and correlate with splunk/k8s. Backed by the metrics MCP tools, which borrow the Brave session cookie and auto-open login when stale. Use when asked to look into metrics, dashboards, or PromQL for a problem or ticket. |
Metrics investigation (metrics)
Query the Grafana metrics backend to investigate a problem (often a Jira
ticket). The metrics MCP tools borrow your Brave session cookie; if the
session is stale the first call auto-opens the Grafana login - finish SSO/MFA
once if prompted, then it proceeds.
1. Make sure metrics is configured
:metrics {:host "metrics.qlikcloud.io"} under :servers in
local-config.edn.gpg, then bb setup.bb, then reload the MCP. macOS + Brave
only.
2. Find the datasource and dashboard
metrics-list-datasources to see the backends (note the Prometheus uid).metrics-search-dashboards with terms from the ticket (service, symptom,
team) to find an existing dashboard.
3. Read what the dashboard already asks
metrics-get-dashboard with the uid dumps each panel's title and PromQL.
Reuse or adapt those expressions instead of guessing.
4. Query and narrow
metrics-query runs a PromQL expression. Default range is the last hour; pass
start/end (e.g. -6h, now, epoch, or RFC3339), step, or
instant=true. Use metrics-metric-names (optionally with a filter
substring) to discover metric names.
5. Correlate
Pull the time window and labels from the metrics, then line them up with logs
(splunk) and, on an SDE, pod state (k8s) for the same service + window. See
the monitor skill for live tailing during a repro.
Notes
- Read-only: every tool is an HTTP GET. No mutation.
- If a call reports "not authenticated", a Brave window was opened - finish
login there and retry.
