Kubeadapt MCP

Kubeadapt MCP connects the AI tools you already use to your Kubeadapt data. Ask Claude why last week cost more than the week before, have Cursor rightsize the Deployment you are editing, or let ChatGPT draft the monthly cost review for your team. The answers come from the same figures you see in the Kubeadapt app.

It is a remote Model Context Protocol server. Your AI tool signs in with your Kubeadapt account; there are no API keys to create or paste.

https://mcp.kubeadapt.io/mcp

Connect your AI tool

AI tool	Where it runs	Setup
Claude	Web, desktop and mobile	Add a custom connector
ChatGPT	Web	Create an app in developer mode
Claude Code	Terminal and IDEs	One command
Cursor	Editor	One-click install
VS Code	Editor, with GitHub Copilot	One-click install
Codex	Terminal and IDEs	Two commands
Gemini CLI	Terminal	One command
Windsurf	Editor	Config file
Zed	Editor	Config file
Other MCP clients	Anywhere	Server URL

What your AI tool can do

Kubeadapt MCP gives your AI tool 23 read-only tools across four kinds of data:

• Costs and carbon: spend by cluster, namespace, workload, node, pod, container, team, department or label; daily, weekly and monthly trends with a month-end estimate; period-over-period comparisons; the split between what workloads use, idle capacity and shared cost; the rates each cluster is priced with; persistent volumes; and carbon emissions.
• Savings: rightsizing, spot, abandoned-workload and orphaned-resource recommendations, with a ready-to-review YAML patch for each workload; capacity and headroom; node and node-group utilization; CPU and memory history; a recommended node layout with what-if estimates; and best-practice findings.
• Alerts: open and recent incidents with their findings and cost impact, and the alert rules and budgets behind them.
• Network costs: traffic cost by billing class, the workloads that send most, internet destinations, cross-zone pairs, and who one workload talks to.

See What you can ask for questions to start with, and the Tools reference for every tool.

Answers you can trust

Every answer carries the same context the Kubeadapt app shows:

• The time window, in UTC calendar days.
• How the figures were obtained: measured from usage, estimated, preliminary (today's data is still arriving), partial or stale.
• The cost mode: fully loaded, which includes each workload's share of idle capacity, or workload only.
• Caveats in plain words, such as a cluster that stopped reporting or traffic with no price, so a figure is never presented as more complete than it is. Missing data is reported as missing, never as zero.
• A link that opens the same view in Kubeadapt.

AI tools are asked to quote the window and the measurement with every number, so the person reading the answer knows what it means.

How access works

1. You add Kubeadapt MCP to your AI tool.
2. The first time it needs data, it opens Kubeadapt in your browser. You sign in as you normally do, including with SSO.
3. You choose what the AI tool may read: which kinds of data and which clusters.
4. The AI tool can then answer questions, within those choices and your own permissions in Kubeadapt.

The connection is read-only: no tool changes anything in Kubeadapt or in your clusters. Rightsizing patches are proposals you review and apply through your own deployment process. Every tool call is recorded, and you can end a connection at any time under Settings → Kubeadapt MCP in the app. See Permissions and security.

Kubeadapt MCP is included in every plan, and every member of your organization can connect.

Next steps

• Connect your AI tool: step-by-step setup for each AI tool.
• What you can ask: questions to start with, grouped by job.
• Permissions and security: what a connection can reach, and the controls admins have.