prometheus-mcp-server MCP — 사용 사례, 설치 & 라이브 데모

왜 쓰나요

핵심 기능

6 focused tools over the Prometheus API
PromQL instant + range queries with custom intervals
Metric discovery and metadata for agents that don't know your label space
Auth options: basic, bearer, mTLS, custom headers

라이브 데모

실제 사용 모습

prometheus.replay ▶ 준비됨

0/0

설치

클라이언트 선택

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "prometheus": {
      "command": "uvx",
      "args": [
        "prometheus-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop → Settings → Developer → Edit Config 열기. 저장 후 앱 재시작.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "prometheus": {
      "command": "uvx",
      "args": [
        "prometheus-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Cursor는 Claude Desktop과 동일한 mcpServers 스키마 사용. 프로젝트 설정이 전역보다 우선.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "prometheus": {
      "command": "uvx",
      "args": [
        "prometheus-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Cline 사이드바의 MCP Servers 아이콘 클릭 후 "Edit Configuration" 선택.

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "prometheus": {
      "command": "uvx",
      "args": [
        "prometheus-mcp-server"
      ],
      "_inferred": true
    }
  }
}

Claude Desktop과 같은 형식. Windsurf 재시작 후 적용.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "prometheus",
      "command": "uvx",
      "args": [
        "prometheus-mcp-server"
      ]
    }
  ]
}

Continue는 맵이 아닌 서버 오브젝트 배열 사용.

~/.config/zed/settings.json

{
  "context_servers": {
    "prometheus": {
      "command": {
        "path": "uvx",
        "args": [
          "prometheus-mcp-server"
        ]
      }
    }
  }
}

context_servers에 추가. 저장 시 Zed가 핫 리로드.

claude mcp add prometheus -- uvx prometheus-mcp-server

한 줄 명령. claude mcp list로 확인, claude mcp remove로 제거.

사용 사례

실전 활용법: prometheus-mcp-server

How to diagnose a latency spike with Prometheus + Claude

👤 On-call SREs ⏱ ~10 min intermediate

언제 쓸까: A service p99 alert fires — you need context without memorizing PromQL.

사전 조건

Prometheus URL reachable — Set PROMETHEUS_URL in the MCP config; add auth if protected

흐름

Scope the spike

Query http request p99 latency for service X in the last hour, 30-second resolution. Compare to the last 7 days baseline.✓ 복사됨

→ Range query result showing the spike
Find correlated metrics

For the spike window, what other metrics for service X moved >2 sigma? CPU, memory, GC, queue depth?✓ 복사됨

→ Candidate culprit metrics
Narrow by label

Break down the spike by pod/host labels. Is it one pod or fleet-wide?✓ 복사됨

→ Per-label decomposition

결과: A hypothesis tied to specific metrics in under 5 minutes.

함정

Query returns no data — Check label names with list_metrics — label casing and delimiters vary between exporters

함께 쓰기: kubectl

Generate a weekly SLO compliance report from Prometheus

👤 SRE leads ⏱ ~25 min intermediate

언제 쓸까: Friday SLO review — you want numbers not vibes.

흐름

Define the SLIs

For service X, compute this week's availability (success/total ratio) and latency (requests under threshold / total) as numbers.✓ 복사됨

→ Two ratios with burn rate
Compare to SLO

Availability SLO = 99.9%, latency SLO = 95%. Am I above or below? Project error budget exhaustion.✓ 복사됨

→ Verdict + days of budget remaining

결과: Defensible SLO report with numbers not 'mostly fine'.

함께 쓰기: google-sheets

Audit Prometheus scrape target health with Claude

👤 Platform engineers ⏱ ~15 min intermediate

언제 쓸까: You suspect half your targets are down but haven't checked.

흐름

Get targets

Call get_targets. Group by job; which have any DOWN instances?✓ 복사됨

→ Table of job → up/down counts
Investigate

For the worst offender, show the lastError for the DOWN instances. Likely cause?✓ 복사됨

→ Actionable cause per target

결과: Rescued scrapes in minutes.

조합

다른 MCP와 조합해 10배 효율

prometheus + kubectl

Pair metric anomalies with pod state

For the service with the latency spike, correlate Prometheus data with kubectl describe on its pods.✓ 복사됨

prometheus + sentry

Metric spike + error spike correlation

Sentry shows errors doubled at 14:00 — what Prometheus metrics moved at the same time?✓ 복사됨

도구

이 MCP가 노출하는 것

도구	입력	언제 호출	비용
health_check		Verify connectivity	1 API call
execute_query	query: promql, time?	Instant snapshot	1 query
execute_range_query	query, start, end, step	Time-series analysis	1 query (may be expensive)
list_metrics	match?: str	Discovery when you don't know the metric name	1 API call
get_metric_metadata	metric: str	Understand units before computing	1 API call
get_targets		Scrape health	1 API call

비용 및 제한

운영 비용

API 쿼터: Prometheus scales with your server; expensive queries can stress it
호출당 토큰: Range queries with many series can hit 10k+ tokens
금액: Free
팁: Use step wisely on range queries; 10s resolution over 24h is 8640 samples per series

보안

권한, 시크릿, 파급범위

최소 스코프: read-only access to Prometheus API

자격 증명 저장: Bearer token or basic auth in env; mTLS cert paths if used

데이터 외부 송신: Your Prometheus URL only

문제 해결

자주 발생하는 오류와 해결

Query returns empty with no error

Metric/label name doesn't exist. Use list_metrics with a match prefix to verify

Range query times out

Reduce time range or increase step. Prometheus query engine has per-query resource limits

401 with bearer token

Token lacks read permission on /api/v1; check reverse proxy if Prometheus is behind one

확인: curl -H 'Authorization: Bearer $T' $PROMETHEUS_URL/api/v1/status/config

대안

prometheus-mcp-server 다른 것과 비교

대안	언제 쓰나	단점/장점
Grafana MCP	You already visualize in Grafana and want dashboard/alert ops	Heavier; more features than you may need
Datadog MCP	Datadog is your metrics store	Paid; different query language

prometheus-mcp-server