devops-mcp-server

A Model Context Protocol (MCP) server that exposes common DevOps/SRE operations — Kubernetes pod status, Prometheus-style metrics queries, and recent deploy history — as tools any MCP-compatible client (Claude Desktop, Claude Code, or a custom LangGraph agent) can call.

Part of a portfolio of agentic AI / AIOps projects. See the index for the full set.

Why this exists

MCP is now supported natively across Anthropic, OpenAI, Google, and Microsoft's agent stacks, but hands-on experience actually building an MCP server (not just consuming public ones) is still uncommon. This repo demonstrates the pattern applied to the domain I know best after 20+ years in infra: exposing operational tooling to an agent in a controlled, typed, auditable way.

Related MCP server: MCP Observability Server

Tools exposed

Each tool ships with a strict input schema and read-only mock data sources — safe to demo without cloud credentials. Swap mock_backends.py for real kubernetes client / prometheus-api-client / GitHub Actions API calls to go to production.

Architecture

flowchart LR
    C[MCP Client<br/>Claude Desktop / Code / custom agent] -->|JSON-RPC over stdio| S[devops-mcp-server]
    S --> K[get_pod_status]
    S --> D[list_recent_deploys]
    S --> M[query_metric]
    K -.-> MB[(mock_backends.py)]
    D -.-> MB
    M -.-> MB

Quickstart

pip install -r requirements.txt
python server.py

Connect it to Claude Desktop

Add to your claude_desktop_config.json (see claude_desktop_config.example.json):

{
  "mcpServers": {
    "devops-tools": {
      "command": "python",
      "args": ["/absolute/path/to/devops-mcp-server/server.py"]
    }
  }
}

Restart Claude Desktop, then ask: "Check pod status for payments-api-deployment/pod-7c9f4."

Project layout

devops-mcp-server/
├── server.py                          # MCP server + tool registration
├── mock_backends.py                   # swappable data sources
├── claude_desktop_config.example.json
├── requirements.txt
└── Dockerfile

Extending this

aiops-agent-orchestrator calls these same tools programmatically (not just from a chat client), which is what turns this from a demo into an actual AIOps building block.

License

MIT