Skip to main content

Tools

The server exposes 25 tools, grouped below by what they do. Each tool needs the capability in its row.

Discovery

ToolWhat it doesCapability
list_telemetry_sourcesEnvironments in the organization and which of metrics, logs and traces each one holds. The first call in any session: every other tool takes an environment name from heretenants:read
list_metricsMetric names an environment holds in a window, optionally narrowed by a search stringmetrics:read
describe_metricOne metric's type, unit and labels with example values, plus PromQL that queries it correctlymetrics:read
list_log_labelsThe labels log streams carry, or one label's values. These are what a LogQL stream selector can filter onlogs:read
list_trace_tagsThe attributes traces carry, grouped by scope, or one attribute's values. These are what a TraceQL search can filter ontraces:read

Query

ToolWhat it doesCapability
query_metricsRun PromQL over a window. Returns exact statistics for every series matched plus a reduced set of pointsmetrics:read
query_logsRun LogQL over a window. Lines newest first, or oldest first to find when something began. An aggregating expression returns a series instead of lineslogs:read
get_traceOne trace's spans as a tree with service, duration, offset and status. A large trace keeps every span that failed, then the slowest, then their parentstraces:read

Charts

ToolWhat it doesCapability
render_chartChart metrics as a Vega-Lite v5 specification with the points inline, plus exact statistics per series and a portal link to the same querymetrics:read

Correlate

ToolWhat it doesCapability
correlate_spanWhich logs and metrics describe one span. It reads the environment's own trace-to-logs and trace-to-metrics configuration, then returns a LogQL selector, any configured PromQL, and the window to run them overtraces:read
find_exemplar_tracesThe traces behind a metric, largest value first. An exemplar is a real trace the store kept while recording a sample, so a latency histogram can hand back a trace that took that longmetrics:read

Dashboards

ToolWhat it doesCapability
list_dashboard_foldersThe folders your dashboards sit indashboards:read
list_dashboardsDashboards with their folder, panel count and last changedashboards:read
get_dashboardOne dashboard's folder, tags, version and every panel with the queries it asksdashboards:read
create_dashboardCreate a dashboard from a list of panels. xScaler works out the layout, the schema and which store each panel queriesdashboards:write
import_dashboardImport an exported JSON document. Both schemas work, classic and resource, and xScaler fills the document's datasource inputs from the environment you namedashboards:write

Alerting

ToolWhat it doesCapability
list_alert_rulesAlert rules and what each is doing now, firing firstalerts:read
get_alert_ruleOne rule's exact state, the queries its condition is built from, what it does on no data, and every series it is firing onalerts:read
list_firing_alertsWhat is firing right now and who is being told, including a branch that routes nowhere and a silence holding an alert backalerts:read
list_contact_pointsWhere alerts can be sent: name, integration, which settings are filled in, and whether anything routes there. No setting value is ever returnedalerts:read
get_notification_policyThe routing tree that decides which contact point an alert reachesalerts:read
check_alert_ruleRun a rule against the data as it is now and save nothing. Answers whether the condition holds, and on which seriesalerts:write
create_alert_ruleCreate a rule from a query and a threshold. xScaler builds the evaluation graph and runs the query once before it saves the rulealerts:write
pause_alert_ruleStop a rule evaluating, or start it again. It writes the paused flag and nothing else, so an edit somebody else is making survives italerts:write

Vocabulary the tools use

Environment. Every tool that touches data takes an environment name, which is the name of a tenant as it appears in the portal under Administration → Tenants. No tool accepts a datasource identifier. Where exactly one environment carries the signal in question, the name can be left out.

Time. Relative first: now-15m, now-1h, now-24h, now-7d. RFC3339 timestamps also work. The default window is the last hour. A time nothing can parse comes back as an error rather than a silent default, and every answer echoes the window and step it used.

Query languages. PromQL for metrics, LogQL for logs, TraceQL for traces.


How to read a tool's answer covers what a tool returns.