Tools Reference

Tools Reference

The server exposes 12 tools across four categories. Your AI assistant selects and calls these automatically.

Log Tools (Loki)

query_pod_logs

Retrieves raw logs from a specific pod or set of pods. All output is automatically sanitized.

Parameter	Type	Default	Description
pod_name	string	required	Pod name or regex pattern
namespace	string	default	Kubernetes namespace
container	string	none	Filter to a specific container
timeframe_minutes	integer	60	Minutes of logs to retrieve
limit	integer	100	Max log lines (capped at 500)

query_logs_custom

Executes a raw LogQL query directly against Loki.

Parameter	Type	Default	Description
query	string	required	Valid LogQL query
timeframe_minutes	integer	60	Time range
limit	integer	100	Max log lines

---

Metrics Tools (Prometheus)

get_pod_metrics

Retrieves a specific metric type for a pod using pre-built PromQL queries.

Parameter	Type	Default	Description
pod_name	string	required	Pod name or regex pattern
namespace	string	default	Kubernetes namespace
metric_type	string	cpu	One of: cpu, memory, restarts, network_rx, network_tx

query_prometheus

Executes a raw PromQL query.

Parameter	Type	Default	Description
query	string	required	Valid PromQL expression
range_query	boolean	false	Get values over time instead of instant
timeframe_minutes	integer	60	Time range for range queries
step	string	1m	Resolution step (e.g., 1m, 5m, 1h)

get_cluster_health

Returns a snapshot of overall cluster health. No parameters required.

Returns: node_count, pod_count, running_pods, failed_pods, pending_pods, cpu_utilization, memory_utilization.

---

Trace Tools (Tempo)

get_trace

Retrieves a complete distributed trace by its trace ID.

Parameter	Type	Default	Description
trace_id	string	required	Hex trace ID, 16–32 characters

search_traces

Searches for traces matching a combination of filters.

Parameter	Type	Default	Description
service_name	string	none	Filter by service name
operation	string	none	Filter by operation/span name
tags	object	none	Filter by span tags as key-value pairs
min_duration	string	none	Minimum trace duration (e.g., 100ms, 1s)
max_duration	string	none	Maximum trace duration
timeframe_minutes	integer	60	How far back to search
limit	integer	20	Max traces to return (max 100)

---

Analytics Tools

analyze_logs

Fetches logs and runs pattern analysis to detect errors, categorize them, and generate recommendations.

Parameter	Type	Default	Description
service_name	string	required	Service or pod name
namespace	string	default	Kubernetes namespace
timeframe_minutes	integer	60	Time window
limit	integer	500	Max log lines to analyze

build_incident_timeline

Builds a unified chronological timeline correlating logs, metrics, and traces.

Parameter	Type	Default	Description
service_name	string	required	Affected service name
namespace	string	default	Kubernetes namespace
timeframe_minutes	integer	60	Time window

enrich_alert

Gathers full context when an alert fires: logs, metrics, traces, and suggested queries.

Parameter	Type	Default	Description
alert_name	string	required	Name of the alert that fired
service_name	string	required	Affected service
namespace	string	default	Kubernetes namespace
timeframe_minutes	integer	30	Context window (max 60)

get_resource_costs

Calculates resource cost attribution by namespace based on CPU and memory usage.

Parameter	Type	Default	Description
namespace	string	none	Filter to a namespace (omit for all)
timeframe_minutes	integer	60	Usage averaging window

Cost estimates use approximate rates ($0.05/core-hour CPU, $0.01/GB-hour memory) for relative comparison only. Refer to AWS Cost Explorer for authoritative figures.

check_slo_status

Checks SLO compliance including error budget remaining and burn rate.

Parameter	Type	Default	Description
service_name	string	required	Service to check
namespace	string	default	Kubernetes namespace
availability_target	float	0.999	Target availability (e.g., 0.999 = 99.9%)
latency_target_ms	float	500.0	Target latency in milliseconds
latency_percentile	float	0.99	Latency percentile (e.g., 0.99 = p99)
window_hours	integer	24	Measurement window in hours (1–720)