Skip to main content

MCP Server Introduction

The CloudYali MCP server is a local, read-only Model Context Protocol server that connects your CloudYali account to an AI assistant. It exposes 35 read-only actions across five areas: cost and usage, budgets, savings recommendations, cost anomalies, and asset inventory. You ask a question in plain English and the assistant answers from your own live CloudYali data, instead of sending you to a dashboard.

It runs on your machine, talks to the CloudYali API over HTTPS, and cannot change anything in CloudYali or in your cloud accounts.

Asking which services cost the most, and the spend answer coming back by cloud

Who it is for​

Anyone who already works in an AI assistant and would rather ask a question than build a report. It suits three situations in particular:

  • Cross-domain questions. "Spend jumped on RDS. Which accounts, and do we have open recommendations against them?" spans cost and recommendations, which are two separate screens in any console.
  • Questions nobody built a screen for. The data is in CloudYali whether or not a view exists for that particular cut of it.
  • Producing an artefact. The assistant holds the actual rows, so it can lay them out as a month-end review, a budget one-pager or an anomaly write-up rather than reciting a number.

What you can ask​

Cost

  • "Break down last month's AWS bill by service."
  • "Which five services cost us the most this month, across every cloud?"
  • "Compare this month's total spend to last month. What changed?"

Budgets

  • "How are we tracking against our budgets?"
  • "Which resources drove the production budget's spend in June?"
  • "Has the data-transfer budget fired any alerts in the last month?"

Savings recommendations

  • "What are our top savings opportunities right now?"
  • "List open recommendations for EBS volumes with more than $50/month savings."
  • "How much could we save in total if we actioned every recommendation?"

Anomalies

  • "Any cost anomalies in the last seven days?"
  • "Summarize anomaly count and impact for the quarter."

Inventory

  • "How many EC2 instances do we have, per region?"
  • "Find all resources tagged environment=prod that are still active."
  • "What did instance i-0abc123 cost last month?"

The full list, with the action behind each one, is in Actions Reference.

Ask for the report, not the number​

Because the assistant receives the underlying rows rather than a rendered chart, it can arrange them. That turns a question into a deliverable.

A month-end review built from one request: the total, the trend, what changed and why

Requests that work well:

  • "Pull August spend, budgets, anomalies and open savings, and build me a review I can send to the leadership team."
  • "Show budget burn as of this morning, worst first, and flag anything over 90%."
  • "Chart daily spend for this account around the 29th, mark the anomaly, and write two paragraphs on what changed."
  • "Group open recommendations by how hard they are, not how much they save."

What it cannot do​

  • It cannot change anything. No creating or editing budgets, no changing recommendation status, no editing tags, no account settings, and no calls at all against AWS, Azure or Google Cloud. See Security and Permissions.
  • It cannot see more than you can. It inherits your own CloudYali permissions exactly.
  • It cannot run in a browser-only assistant. The server is a local process, so ChatGPT on the web and claude.ai in a tab cannot reach it. Desktop and terminal clients work today.

Two things to keep in mind​

An empty result is not proof of zero spend. A filter that matched nothing and a period with no cost look identical in a chat reply. If a number surprises you, ask the assistant which filters it applied.

Spot-check any number you are about to quote to someone else. The data is identical to the console, but the assistant's choice of grouping, date range and cost types may not match the view you have in your head.

Next steps​

Last updated on