Ask Claude why production is down. Get answers from your live systems.

devops-mcp lets Claude read your Kubernetes, Jenkins, AWS and other DevOps systems, so on-call engineers ask one question instead of checking six dashboards.

pipx install "git+https://github.com/gerardrecinto/devops-mcp.git"
claude mcp add devops-mcp -- devops-mcp
devops-mcp --check
Browse the tools

Needs Python 3.11 or later, pipx on your PATH, and Claude Code to register the server. It installs from GitHub: the devops-mcp name on PyPI is a different project.

MCP is a standard plug that lets Claude use outside tools, and this one lets it look at your servers and build systems.

33 tools9 servicesRead-only by default

Then ask a question

With no setup it uses your current kubeconfig. Ask Claude "Which pods are failing in my cluster?" and it answers about your real cluster. devops-mcp --check shows which services have local configuration. It does not test API connectivity or credentials.

You: "The API deployment has been stuck for 20 minutes. What's going on?"

Claude:
  [k8s_get_failing_pods("production")]
  api-server-7d9f8b-xk2p9  CrashLoopBackOff  restarts: 14

  [k8s_pod_logs("api-server-7d9f8b-xk2p9", tail_lines=30)]
  [09:10:03] FATAL  postgresql://db-prod:5432: Connection refused

  [k8s_get_events("production", event_type="Warning")]
  OOMKilling  Pod/db-prod-0  "Memory limit exceeded"  count: 3

  db-prod-0 was OOMKilled, so the API pods cannot reach the database.

illustrative Pod names, incident number and times are made up. This is not output from a real system.

The tools

Built from the source code on every deploy.

Each card is one question Claude can ask a system for you, such as which builds are failing.

Loading the tool list...

Connect your systems

Put credentials for the services you use in ~/.devops-mcp/.env (copy .env.example from the repo), then restart Claude. Services without credentials are skipped.

Build a starter .env

This makes a text file of the logins Claude needs, with blanks for you to fill in.

Services

~/.devops-mcp/.env

Replace every placeholder and delete the lines for services you do not use. Nothing here is sent anywhere. Kubernetes needs no entry.

Docker, extras, Claude Desktop and sample data

Azure, GCP and Slack need extra packages:

pipx install --force "mcp-devops[azure,gcp,slack] @ git+https://github.com/gerardrecinto/devops-mcp.git"

Docker (the image includes every integration):

docker pull ghcr.io/gerardrecinto/devops-mcp:latest
claude mcp add devops-mcp -- docker run -i --rm --env-file ~/.devops-mcp/.env \
  -v ~/.kube:/home/devops-mcp/.kube:ro ghcr.io/gerardrecinto/devops-mcp:latest

Claude Desktop: add this to ~/Library/Application Support/Claude/claude_desktop_config.json and restart.

No credentials yet? The demo flag returns sample data. Every response is labelled as sample data and is not real information about your systems. Remove the flag once you connect real ones.

claude mcp add devops-mcp -e DEVOPS_MCP_DEMO=1 -- devops-mcp

What it does and does not do

Claude can look at your systems but not change them, apart from chat messages you switch on, and a private log records every look.

  • Read-only by default. It cannot apply, delete, restart or trigger anything. The only writes are Slack and Teams messages, and they stay off unless you set DEVOPS_MCP_ENABLE_WRITES=1.
  • Every call is logged to ~/.devops-mcp/audit.log, with secret-like values hidden.
  • It reports what the APIs return. Claude's guess at a root cause is a hypothesis to check, not a verdict.
  • No live demo here. The tools need your credentials and systems, so this page shows the catalog.