Skip to content
Start free

Kubeadapt MCP

Review monthly Kubernetes spend

Ask an AI tool connected to Kubeadapt MCP for a monthly Kubernetes cost review: spend by cluster or namespace, the change, and a month-end estimate.

Last updated

Ask your AI tool what Kubernetes cost this month and it answers with the total, the change on the previous period, the clusters or namespaces that cost the most, and an estimate of where the month will end. The figures are the ones the Kubeadapt Cost Explorer shows, with the window and cost mode stated in every answer.

This example needs a connection with the Costs and carbon data kind. The budget question also needs Alerts.

Ask

plaintext
What did we spend on Kubernetes this month so far, by cluster? How does that compare with the same number of days before?
plaintext
Break down last month's spend on the prod cluster by namespace. Which five namespaces cost the most, and what share of the total is each?
plaintext
Where will this month's spend end up if nothing changes?
plaintext
Show daily spend for the whole organization over the last 30 days. Which days stand out?
plaintext
Which budgets are we close to, and how much of each is spent?
plaintext
Draft a one-page monthly cost summary for my manager: total, change, top five namespaces and the month-end estimate. Quote the window and cost mode.

What happens

ToolWhat it returns here
get_contextYour organization, the clusters the connection can see with their ids and monthly run rate. AI tools call it first to turn "the prod cluster" into an id.
query_costsSpend for the window grouped by cluster, namespace or workload, the previous period of equal length and the change, the top rows with their share and run rate, and cost_split.
get_cost_trendSpend per day, week or month for the organization, a cluster, a namespace or a workload, and month_end_projection when the window covers the current month.
compare_costsTwo periods side by side, when you name both, such as "September against August".
list_alert_rulesBudgets with their amount, period and alert percentages, and the share of this period's budget spent when a budget alert has recorded it.

What the answer contains

  • Total and change. total_usd for the window, previous_period_usd for the period of equal length just before it, and change_pct. "This month so far" compared with "last month" is month to date against the same number of days before, unless you give both periods.
  • Rows. Each cluster, namespace or workload with cost_usd, share_of_total_pct, run_rate_usd_per_month, its CPU, memory, GPU and network parts, and an id the AI tool can open with fetch. Answers list the top 25 rows by default; the total always covers every row.
  • Where the money sits. cost_split divides the fully loaded bill into workload usage, idle capacity nobody requested, and shared and unallocated cost (storage, the cluster management fee and other cost no single workload owns).
  • Month-end estimate. Spend on the complete days of the month so far plus the average of the last 7 complete days for each day left. The answer labels it as an estimate, not a forecast, and leaves it out when fewer than 3 recent days have data.
  • Provenance. The window in UTC calendar days, the cost mode, and whether the figures are measured, preliminary (the window includes today), partial or a lower bound.

Cost mode. Answers use fully loaded cost unless you ask otherwise: each row carries its share of idle capacity, and shared cost sits in a separate Shared & Unallocated row. Ask for workload-only cost to see what each row uses or requests on its own. Grouped by node, totals cover the node bill only, without storage or the management fee, so they sit below the cluster total. See Cost modes.

Example (illustrative)

The shape of a typical answer, without real figures:

plaintext
Spend from 2026-10-01 to 2026-10-12 (fully loaded): [total], [change] on the previous 12 days.
Largest clusters: prod-eu [share], prod-us [share], staging [share].
Of the total, [amount] is idle capacity and [amount] is shared and unallocated.
October is on track for about [estimate] (estimated from the last 7 complete days).
Measured; 1 cluster stopped reporting and is left out. Open in Kubeadapt: [link]

Follow-ups

plaintext
Show the same breakdown in workload-only cost.
plaintext
Within the most expensive namespace, which workloads cost the most?
plaintext
Which rates is the prod cluster priced with? Are any clusters on Kubeadapt's default rates?
plaintext
What was the carbon footprint of each cluster over the same window?

Limits

  • Read-only. Kubeadapt MCP reports spend; it changes nothing in Kubeadapt or in your clusters.
  • One window per question, up to 366 days. Dates are UTC calendar days, and the end date can be today at the latest. A window that includes today is preliminary, because today's data is still arriving.
  • Up to 200 rows per answer. Ask for a filter, such as a cluster or namespace, to see rows beyond that.
  • Labels filter, they don't group. Kubeadapt can show the cost of one label value at a time. To compare a label's values, the AI tool asks once per value; see Team chargeback.
  • US dollars. Money in every answer is in US dollars.
  • Your clusters only. Answers cover the clusters the connection can see, within your own permissions. A question can name up to 50 clusters.

Related