Signoz

SigNoz MCP

Query a live SigNoz instance over MCP (server already configured). ~41 signoz_* tools across telemetry (metrics/traces/logs), alerts, dashboards, saved views, notification channels, and docs -- plus 4 built-in workflow prompts.

This skill is for reading and operating a running SigNoz. For designing SLOs, alert thresholds, or dashboards from first principles use observability-designer. For reading/acking incidents use the regen agent -- SigNoz is where you land after extracting an incident's correlation keys.

Two rules that make queries fast (and correct)

1. Filter by resource attributes first. Resource attributes (service.name, k8s.namespace.name, host.name) are indexed and dramatically speed up backend queries. If you don't already have one:

signoz_get_field_keys(signal="logs", fieldContext="resource")   # discover keys
signoz_get_field_values(signal="logs", name="service.name")     # discover values

Then always scope queries with a service (or filter: service.name = '...'). Do not run broad, unfiltered log/trace scans.

2. Pick the operator to match intent AND data type.

Intent	Operator	Example
Field exists	EXISTS	trace_id EXISTS
Field absent	NOT EXISTS	k8s.pod.name NOT EXISTS
Exact match	=	service.name = 'frontend'
Exclude (field must exist)	EXISTS AND !=	service.name EXISTS AND service.name != 'redis'
One of several	IN	severity_text IN ('ERROR','WARN','FATAL')
Substring	LIKE / ILIKE	body ILIKE '%timeout%'
Containment	CONTAINS	body CONTAINS 'timeout'
Regex	REGEXP	name REGEXP '^grpc\.'

Safe operators by type: bool = != EXISTS; int64 = != > >= < <= IN EXISTS; string all of the above plus LIKE ILIKE CONTAINS REGEXP IN NOT IN.

Caveat: negative operators (!=, NOT LIKE, NOT IN) only match rows where the field is present -- rows missing the field are silently dropped. To include them, OR in a NOT EXISTS, or gate with EXISTS AND != when you mean "present but not X".

Signal ambiguity: if it isn't clear whether the answer lives in metrics, traces, or logs, ask before querying.

The mini-axol / Riddler deployment

The configured signoz MCP points at a self-hosted SigNoz on the tailnet:

• signoz_url = http://mini-axol.tail9b2ce8.ts.net:3301 -- Tailscale must be up (a connection failure is almost always a down tailnet, not a SigNoz outage).
• Telemetry source is Riddler OTel; custom metrics are named riddler_*.
• Primary filter is always service.name. Start any investigation with signoz_list_services to confirm reachability and the live service names.

Tool map (~41 tools)

Full argument detail in tools-reference.md; workflows in usage-patterns.md.

Group	Tools
Discovery	signoz_list_services, signoz_get_service_top_operations, signoz_list_metrics, signoz_get_top_metrics, signoz_get_field_keys, signoz_get_field_values
Metrics	signoz_query_metrics, signoz_execute_builder_query, signoz_check_metric_cardinality, signoz_check_metric_usage
Logs	signoz_search_logs, signoz_aggregate_logs
Traces	signoz_search_traces, signoz_aggregate_traces, signoz_get_trace_details
Alerts	signoz_list_alerts, signoz_list_alert_rules, signoz_get_alert, signoz_get_alert_history, signoz_create_alert, signoz_update_alert, signoz_delete_alert
Dashboards	signoz_list_dashboards, signoz_get_dashboard, signoz_create_dashboard, signoz_update_dashboard, signoz_delete_dashboard, signoz_import_dashboard, signoz_list_dashboard_templates
Saved views	signoz_list_views, signoz_get_view, signoz_create_view, signoz_update_view, signoz_delete_view
Notification channels	signoz_list_notification_channels, signoz_get_notification_channel, signoz_create_notification_channel, signoz_update_notification_channel, signoz_delete_notification_channel
Docs	signoz_search_docs, signoz_fetch_doc

Built-in MCP prompts

The server ships 4 slash-command prompts (they surface as /signoz:<name>). They are ready-made workflows, not clutter -- reach for them first when they fit:

Prompt	Args	Runs
/signoz:debug_service_errors	service, timeRange	error logs + error-trace aggregate + top operations
/signoz:latency_analysis	service, timeRange	p99 latency metrics + trace-duration aggregate + slow ops
/signoz:compare_metrics	metricName, period1, period2	one metric across two windows to spot regressions
/signoz:incident_triage	alertId	alert details + history + related logs & traces

These prompts cannot be individually hidden in Claude Code (no supported toggle); see usage-patterns.md for the optional build-time opt-out and why keeping them is recommended.

What You Get

• Reference documentation for all ~41 SigNoz MCP tools grouped by signal and function.
• The resource-attribute-first rule and a filter-operator cheat sheet that prevent slow scans and silently-dropped rows.
• Concrete workflows including the regen incident -> OTel correlation loop against the mini-axol/Riddler deployment.

See also

• regen -- reads Fluidify Regen incidents and emits the service.name + labels + time window to feed into these tools (the incident side of the loop; see usage-patterns.md)
• observability-designer -- design SLOs, alert rules, and dashboards conceptually
• ethskills -- Ethereum tooling context for the Riddler stack being observed