Skip to content

Instantly share code, notes, and snippets.

@tjhop
Created March 7, 2026 05:58
Show Gist options
  • Select an option

  • Save tjhop/5390e3a205ea5f5973ecffb03a5bb5f1 to your computer and use it in GitHub Desktop.

Select an option

Save tjhop/5390e3a205ea5f5973ecffb03a5bb5f1 to your computer and use it in GitHub Desktop.
Helm chart install and connecting claude code over HTTP
▐▛███▜▌ Claude Code v2.1.71
▝▜█████▛▘ Opus 4.6 · Claude Max
▘▘ ▝▝ ~/go/src/github.com/tjhop/prometheus-mcp-server
❯ I'm new to this prometheus instance, can you give me an overview?
● Let me gather information about your Prometheus instance.
● prometheus - Health Check (MCP)
⎿  Prometheus Server is Healthy.
● prometheus - Build Info (MCP)
⎿  {
"version": "3.10.0",
"revision": "54e010926b0a270cadb22be1113ad45fe9bcb90a",
… +5 lines (ctrl+o to expand)
● prometheus - Runtime Info (MCP)
⎿  {
"startTime": "2026-03-07T05:40:39.736983714Z",
"CWD": "/prometheus",
… +9 lines (ctrl+o to expand)
● prometheus - TSDB Stats (MCP)
⎿  {
"headStats": {
"numSeries": 30594,
… +41 lines (ctrl+o to expand)
● prometheus - List Targets (MCP)
⎿  Error: result (1,232,421 characters) exceeds maximum allowed tokens. O
utput has been saved to /home/tjhop/.claude/projects/-home-tjhop-go-sr
c-github-com-tjhop-prometheus-mcp-server/867954b7-7947-41a7-b778-5da7c
… +14 lines (ctrl+o to expand)
● prometheus - List Alerts (MCP)
⎿  {
"alerts": [
{
… +76 lines (ctrl+o to expand)
● Here's an overview of your Prometheus instance:
Server
┌────────────┬─────────────────────────────────────┐
│ │ │
├────────────┼─────────────────────────────────────┤
│ Version │ Prometheus 3.10.0 (Go 1.26.0) │
├────────────┼─────────────────────────────────────┤
│ Started │ 2026-03-07 05:40 UTC (today) │
├────────────┼─────────────────────────────────────┤
│ Health │ Healthy, config loaded successfully │
├────────────┼─────────────────────────────────────┤
│ Retention │ 10 days │
├────────────┼─────────────────────────────────────┤
│ GOMAXPROCS │ 16, GOGC: 75 │
└────────────┴─────────────────────────────────────┘
TSDB Stats
- ~30,600 active series across 1,422 unique metrics
- 3,625 label pairs
- Top metrics by series count are mostly apiserver histogram buckets
(apiserver_request_body_size_bytes_bucket,
apiserver_request_duration_seconds_bucket,
etcd_request_duration_seconds_bucket)
Environment
This looks like a Kubernetes cluster monitored via kube-prometheus-stack. The
dominant label pairs confirm this:
- job=apiserver (20,659 series)
- namespace=default (20,672 series)
- namespace=prometheus (5,155 series)
- namespace=kube-system (4,120 series)
The targets list is large (1.2M characters), suggesting many scrape targets
are configured.
Active Alerts (8 firing/pending)
Alert: TargetDown (x4)
State: firing
Summary: kube-scheduler, kube-controller-manager, kube-etcd, kube-proxy -- all
100% down in kube-system
────────────────────────────────────────
Alert: Watchdog
State: firing
Summary: Expected -- dead man's switch for alerting pipeline
────────────────────────────────────────
Alert: PrometheusNotConnectedToAlertmanagers
State: firing
Summary: Prometheus has no Alertmanager connections
────────────────────────────────────────
Alert: etcdMembersDown
State: pending
Summary: 1 etcd member down
────────────────────────────────────────
Alert: etcdInsufficientMembers
State: firing
Summary: 0 sufficient etcd members (critical)
Key Observations
1. This appears to be a local/dev cluster (likely kind or minikube) -- the
TargetDown alerts for kube-scheduler, kube-controller-manager, kube-etcd, and
kube-proxy are common in kind clusters where those components aren't exposed
as separate scrape targets.
2. Prometheus is not connected to any Alertmanagers, so alerts won't be routed
anywhere.
3. Cardinality is modest at ~30K series -- no concerns there.
Want me to dig into anything specific -- targets, particular metrics, the
alerts, or something else?
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment