TL;DR
The DevOps Stack is 6 MCPs (GitHub, Docker, AWS, Sentry, Grafana, Slack) that cover CI/CD, cloud, observability, and incident response. One Claude session handles deploys, debugs, and incidents end-to-end. Replaces ~$400/mo of disparate SaaS subscriptions and cuts incident response time by 50%.
The DevOps Automation Stack
CI, deploy, observability, and infrastructure — automated end-to-end
Install the whole stack
$ mcpizy install github docker aws sentry grafana slackOne command installs and configures all 6 MCPs for Claude Code, Cursor, Windsurf, or any MCP-compatible client.
Why this stack?
DevOps engineers manage six tools at once, all day: git + CI, containers, cloud, errors, metrics, comms. This stack is those six as MCPs — one Claude session for the whole ops workflow. When a 3am PagerDuty alert fires, you're not re-learning each CLI under pressure; you're describing what you see and Claude is pulling from every MCP to give you a full picture in 30 seconds.
This isn't a replacement for your existing monitoring — it's an amplifier. Grafana + Sentry + CloudWatch are already great; the MCP layer makes them conversational, so the right data shows up without you context-switching.
MCPs in this stack (6)
What this stack lets you do
Container deploy on push
See recipe- 1Push to main — GitHub event
- 2Docker MCP builds + tags image
- 3Image pushed to ECR via AWS MCP
- 4ECS service updated with new task definition
- 5Slack notification posted
3am incident response
See recipe- 1PagerDuty fires — 'API latency spiked'
- 2Claude pulls Grafana for the window + top-10 slow endpoints
- 3Sentry checked for new errors
- 4Docker MCP inspects container CPU on affected hosts
- 5Postgres MCP queries for runaway SQL
- 6Claude identifies root cause, posts status to #incidents in Slack
- 7Linear ticket created for tomorrow's fix
Cost optimization sweep
- 1Claude queries AWS MCP for last-30-day spend by service
- 2Flags top anomalies (idle RDS, oversized EC2, unused EBS)
- 3Cross-references Grafana for utilization metrics
- 4Drafts optimization proposal in Notion + Linear tickets
- 5Slack summary posted to ops channel
Estimated value
Typical DevOps team of 3: ~$400/mo saved on tooling (Datadog add-ons, Runbook.io, operator seats). More importantly: 6–10h/week per engineer reclaimed, 50% reduction in incident mean-time-to-mitigation.
Frequently asked questions
Is it safe to give Claude write access to AWS?
Scope the IAM role — read-only for most work, with a separate role for scoped write (e.g., only Lambda updates, no IAM changes). Most teams start read-only for a week, then widen carefully. Treat the MCP like a junior engineer.
Can Claude actually run kubectl / terraform apply?
Yes (via Kubernetes MCP / Terraform MCP). Most teams gate writes behind a confirmation or restrict to non-prod. For prod, GitOps is safer: Claude opens a PR to the manifest repo, humans merge.
Does this replace Datadog or Sentry?
No — it complements them. Datadog/Sentry/Grafana stay as your source of truth. The MCP layer makes them conversational, so queries are instant instead of 5 clicks.
What about Kubernetes — should I add a Kubernetes MCP?
Yes if you run K8s. Community Kubernetes MCP covers pod inspection, logs, rolling restarts, and kubectl parity. Add it to this stack if you're on K8s — it's the 7th MCP for K8s shops.
Can I use this stack for on-call rotation?
Add PagerDuty MCP (community) for rotation management + incident ack. The 6 above + PagerDuty give you a complete on-call toolkit, all callable from Claude when you're half-asleep at 3am.