I built 3 MCP servers so I can ask Claude about my DevOps stack

Every time something looked off in production, I'd switch between 4 tabs:
Prometheus → check metrics, kubectl → check pods, Grafana → check dashboards, terminal → check logs.

So I built MCP DevOps Pack — 3 MCP servers that let Claude Desktop talk to your infra directly.

## What's included

| Package | What it does |
|---------|-------------|
| @peachjed/mcp-prometheus | PromQL queries, firing alerts, rule inspection |
| @peachjed/mcp-kubernetes | List pods, get logs, describe resources, watch events |
| @peachjed/mcp-grafana | Search dashboards, list datasources, check alert states |

## Install

bash
  npm install -g @peachjed/mcp-prometheus @peachjed/mcp-kubernetes @peachjed/mcp-grafana

  Configure Claude Desktop

  Add to your claude_desktop_config.json:

  {
    "mcpServers": {
      "prometheus": {
        "command": "mcp-prometheus",
        "env": { "PROMETHEUS_URL": "http://localhost:9090" }
      },
      "kubernetes": {
        "command": "mcp-kubernetes"
      },
      "grafana": {
        "command": "mcp-grafana",
        "env": {
          "GRAFANA_URL": "http://localhost:3000",
          "GRAFANA_TOKEN": "your-token"
        }
      }
    }
  }

  What you can ask Claude

  - "What's the current CPU usage across all nodes?"
  - "Show me the last 50 lines from pod api-server-xyz in production"
  - "Are there any firing alerts right now?"
  - "List all dashboards in the Infrastructure folder"

  How it works

  Each server is a small TypeScript process that runs locally via stdio. Claude Desktop spawns it automatically when
  needed. The Kubernetes server uses your existing ~/.kube/config — no extra auth setup.

  Stack

  - TypeScript + @modelcontextprotocol/sdk
  - @kubernetes/client-node for the k8s server
  - Prometheus and Grafana via their HTTP APIs

  Source

  GitHub: https://github.com/Jedsadakorn-Suma/mcp-devops-pack

  npm: @peachjed/mcp-prometheus, @peachjed/mcp-kubernetes, @peachjed/mcp-grafana

  Feedback welcome — especially if you use a different observability stack.