TL;DR
DevOps engineers context-switch across Docker, Kubernetes, AWS, Terraform, and GitHub Actions a hundred times a day. MCPs collapse those into a single Claude Code session where the agent reads logs, rolls back deploys, checks cluster health, and writes the postmortem.
The MCP stack for DevOps, SRE, and platform engineering
A DevOps / SRE / platform engineer running Kubernetes, CI/CD, observability, and on-call rotations. Lives in a terminal.
What hurts today
- 1Jumping between kubectl, aws CLI, terraform plan output, and GitHub Actions logs breaks flow
- 2On-call incidents at 3am mean re-learning each CLI under stress — human error time
- 3Writing runbooks and postmortems is the most dreaded 2h of every incident
- 4Cost optimization (AWS bill, k8s right-sizing) requires 3 different dashboards and a spreadsheet
- 5Terraform drift goes undetected because nobody runs `terraform plan` on unchanged stacks
Recommended MCPs (7)
Docker
View MCPInspect containers, tail logs, restart services, and build images — from Claude Code. Works on your local Docker, a remote host, or a CI runner.
AWS
View MCPS3, Lambda, ECS, CloudWatch — all as MCP tools. 'Show me the last 20 errors in production Lambda X' becomes a one-liner instead of a CloudWatch Insights query.
GitHub
View MCPTrigger Actions, inspect workflow runs, and debug failed jobs without opening the GitHub UI. Critical for fixing CI breakages in the terminal.
Postgres
View MCPDirect DB access for on-call debugging — with read-only mode by default. 'Why is user X stuck?' → Claude queries the DB, finds the row, explains the state.
Grafana
View MCPQuery metrics, list active alerts, and build dashboards via Claude. Especially powerful during incidents: 'show me every spike in 95p latency in the last hour'.
Slack
View MCPAuto-post incident updates, deploy announcements, and daily summaries. Claude writes the status update so you don't have to context-switch from the fix.
Sentry
View MCPTriage errors, group them, assign them to the right service owner — all via Claude. Turns the Sentry inbox into a 5-minute task instead of 45.
A real workflow
3am PagerDuty alert: API latency spiked. You open Claude Code on your phone (via Termux). One prompt: 'investigate the latency spike'. Claude queries Grafana for the spike window, pulls the top 10 slow endpoints, checks Sentry for new errors, inspects Docker container CPU on the affected host, identifies a runaway query, lists the top Postgres statements, kills the bad query, posts an update to #incidents in Slack, and creates a Linear issue tagged @platform for tomorrow's fix. You're back in bed in 8 minutes.
Time ROI
DevOps engineers save 6–10h/week. Biggest wins: incident response (−50% time to mitigation), postmortem writing (−60%), cost optimization (finally actually done).
Recommended recipes for this role
Container Deployment to AWS
Build, tag, push to ECR and deploy to ECS in one automated pipeline triggered by a git tag or manual dispatch.
Database Monitoring Dashboard
Stream Postgres metrics — query latency, lock waits, vacuum stats — into Grafana for a live operations dashboard.
Alert Routing from Grafana
Grafana alerts are enriched with runbook links and routed to the correct Slack channel based on severity and team labels.
Error Alerting Pipeline
Sentry new issues are de-duplicated, enriched with commit info, and routed to the right Slack channel based on project.
Database Change Alerts
Get a Slack alert whenever a critical Supabase table row is created, updated, or deleted — ideal for audit trails.
Preview Deploy on Every PR
Open a PR and a Vercel preview URL appears as a comment within minutes. Branches are cleaned up automatically when PRs close.
Frequently asked questions
Is it safe to give Claude write access to my AWS account?
Scope the IAM role the AWS MCP uses — read-only for most tasks, scoped write (e.g., only Lambda updates, no IAM changes) for routine ops. Treat the MCP like a junior engineer: start read-only for a week, then widen.
Can Claude actually run `kubectl apply` or is that too dangerous?
The Kubernetes MCP exposes both read and write tools. Most teams keep writes gated behind a confirmation prompt or restricted to a staging cluster. For prod, use GitOps (Claude opens a PR to the manifest repo, humans merge).
What about Terraform — can Claude plan and apply?
Yes, via the Terraform MCP (community-maintained). `terraform plan` is safe by default. For `apply`, restrict to non-prod workspaces or require a PR-based workflow where Claude drafts the change and a human approves.
How does this replace or complement my existing runbook tooling?
MCPs make runbooks executable. Instead of 'step 3: check CloudWatch for error rate', your runbook becomes a Claude prompt and Claude runs steps 1–10 itself. You still keep the runbook as the spec for what should happen.
Any MCPs specifically for on-call rotation management?
PagerDuty and OpsGenie both have community MCPs. Useful for 'who's on call right now?', 'acknowledge alert X', 'escalate to secondary'. Pair with Slack MCP for incident channel coordination.
Other use cases
MCPs for SaaS Founders
A technical founder (0–10 employees) building a B2B SaaS who ships code, handles billing, writes marketing, and answers support — all in the same day.
MCPs for Solopreneurs & Indie Hackers
An indie hacker with a Twitter audience, a newsletter, 1–3 shipped products, and zero employees. Ships daily, markets constantly, avoids meetings.
MCPs for AI Agent Developers
A developer building AI agents, chatbots, or autonomous workflows. Needs search, scraping, vector storage, and LLM orchestration — all as tools the agent can call.
Start with this MCP stack
Install the full stack in one command, or cherry-pick the MCPs you need.
Browse all MCPs