Ask FinOps Agent⌘IPricingSign inGet started
Product DocsDevelopers

MCP

Ask it this

Prompts that work as written against the LevelFour MCP server, the tools each one lands on, and the questions the server cannot answer.

Every prompt below is written to be copied as it stands. Each one says which tools it lands on and the mistake it is worded to avoid, because the difference between a useful answer and a confident wrong one is usually a detail the model was never told to check.

Substitute your own service names, months and ids. Read levelfour://filters/costs first if you want the values your data actually contains.

Start here

Using LevelFour, what did we spend in July 2026, and how did that compare
with June?

Lands on get-cost-summary with period="2026-07". One completed month and its change on the month before, which is the comparison that is safe to read as a number.

Today's month is partial. A model that compares a four-day month against a full one will report a collapse in spend. Say "July" rather than "this month" when you want a number to hold still.
Read the quantity column, not the cost column, when you are judging whether something is idle. Cost rounds to cents, and a rounded zero looks exactly like a resource nobody is using.

By question

Longer analyses

The five prompts run a full ordered workflow, not a single call. In Claude Code:

/mcp__levelfour__monthly_bill_review 2026-07
/mcp__levelfour__quarterly_commitment_review
/mcp__levelfour__waste_by_team team
/mcp__levelfour__savings_review REC-1234
/mcp__levelfour__approvals_review

See Prompts and resources.

What it cannot answer

A model chooses its own tools, so the same question asked twice can take a different route. Expand the tool call in your client to see which ran and with what arguments. That is where the two most common wrong answers are visible: an empty result read as a real zero, and a month-over-month comparison run on a month that is four days old. Reading a result covers the fields that qualify an answer.

Next

  • Reading a result is what comes back: the pagination envelope, truncation, limits and errors
  • Tools is the generated reference, with every parameter, enum and default
  • CLI has the write commands this server does not expose

Prompts and resources

The prompts and resources the server ships alongside its tools, so a model stops guessing at method and at values.

API Reference

Where the LevelFour REST API lives, the Bearer token every request carries, the envelope every response comes back in, and the endpoint reference by category.

On this page

Start hereBy questionLonger analysesWhat it cannot answerNext
Ask the FinOps Agent about your cloud spend