kubesense-mcp

Original：🇺🇸 English

Translated

How to use KubeSense MCP tools to query logs, traces, and metrics from Kubernetes clusters. Covers tool selection, the discovery-first workflow, and links to datasource-specific skills.

2installs

Sourcekubesense-ai/kubesense-mcp-skills

Added on2026-08-13

NPX Install

npx skill4agent add kubesense-ai/kubesense-mcp-skills kubesense-mcp

SKILL.md Content

View Translation Comparison →

KubeSense MCP

KubeSense MCP provides observability tools for querying logs, traces, and metrics from Kubernetes clusters.

Tools

Tool	Purpose
`get-trace-or-log-fields`	Discover available fields for logs or traces
`get-available-metrics`	Discover available metric names
`get-metric-labels`	Get label names for a specific metric
`search-logs`	Browse raw log records (max 10 rows)
`search-traces`	Browse raw trace/span records (max 10 rows)
`analyze-logs`	Run aggregated analysis on logs
`analyze-traces`	Run aggregated analysis on traces
`analyze-metrics`	Execute PromQL queries on metrics
`analyze-telemetry`	Multi-datasource queries with formula support

Discovery-First Rule

Always discover before querying. Never guess field names or metric names.

logs/traces:  get-trace-or-log-fields  →  search-logs / search-traces / analyze-logs / analyze-traces
metrics:      get-available-metrics    →  get-metric-labels  →  analyze-metrics

Choosing the Right Tool

Show me recent logs →
```
search-logs
```
How many errors in the last hour? →
```
analyze-logs
```
with
```
row_count
```
+ error filter
P99 latency for a service →
```
analyze-traces
```
with
```
p99
```
aggregation on
```
duration
```
CPU / memory usage →
```
get-available-metrics
```
→
```
analyze-metrics
```
Error rate as a percentage →
```
analyze-telemetry
```
with a formula query

Query Types

All

analyze-*

tools accept a

queryType

:

```
"range"
```
— time-series values over the given window (for trends)
```
"instant"
```
— single point-in-time snapshot (for totals/counts)

Time Ranges

All tools require

from_time

and

to_time

as ISO 8601 strings, e.g.:

"from_time": "2026-04-23T10:00:00Z"
"to_time":   "2026-04-23T10:30:00Z"

Start narrow (15–30 min) and widen if needed.

Filters

The

filters

parameter in

search-logs

,

search-traces

,

analyze-logs

, and

analyze-traces

accepts a SQL-like WHERE clause.

Operators

Operator	Example	Notes
`=`	`level = 'ERROR'`	Exact match
`!=`	`status != 'ok'`	Not equal
`>` `<` `>=` `<=`	`duration_ms > 500`	Comparisons; use unquoted numbers for float fields. Filter trace latency via `duration_ms` (ms), not the ns `duration` field
`IN`	`level IN ('ERROR', 'WARN')`	Matches any value in list
`NOT IN`	`namespace NOT IN ('kube-system')`	Excludes values in list
`LIKE`	`workload LIKE '%api%'`	`%` is wildcard
`NOT LIKE`	`pod_name NOT LIKE '%debug%'`	Negative pattern match

Combining Conditions

level = 'ERROR' AND namespace = 'production'

status = 'error' AND (workload = 'api-server' OR workload = 'auth-service')

Value Types

Strings — always single-quoted:
```
workload = 'my-service'
```
Numbers — unquoted:
```
duration_ms > 500
```
(filter trace latency in ms via
```
duration_ms
```
)

Examples

level IN ('ERROR', 'FATAL') AND workload LIKE '%payment%'

status = 'error' AND protocol_type = 'HTTP' AND return_code = '500'

duration_ms > 500 AND workload = 'checkout-service'

Datasource Skills

For detailed field references, examples, and query patterns, read the datasource-specific skill:

kubesense-logs — Discover fields, search raw logs, aggregate with counts/percentiles, filter syntax
kubesense-apm — Discover fields, search raw spans, analyze latency and errors, filter syntax
kubesense-metrics — Discover metrics, get labels, write PromQL queries

Multi-Query Reference

multi-query.md — Multi-datasource queries and formula expressions with
```
analyze-telemetry
```