Cloud cost data for your AI assistant

An MCP server that gives Claude, Cursor or any MCP client verified cloud rates and the same cost models behind the calculators on this site — with the vendor source and verification date attached to every answer.

No account. No API key. No email.

Install

npx @finitizer/finops-mcp

Claude Code

claude mcp add finitizer-finops -- npx -y @finitizer/finops-mcp

Claude Desktop — claude_desktop_config.json

{
  "mcpServers": {
    "finitizer-finops": {
      "command": "npx",
      "args": ["-y", "@finitizer/finops-mcp"]
    }
  }
}

It reaches nothing

No network calls at all — every rate and model is compiled into the bundle, so it works offline. It holds no customer data and cannot reach an account. Finitizer’s authenticated MCP server is a separate product that shares no code with this one.

Every answer is attributable

Each response carries the vendor URL the rate came from, the date a human last read it, and a link to the page showing the working. An assistant can tell you how current a figure is instead of guessing.

Same models as the site

These are the engines behind the calculators here, not a simplified copy — including the parts that make the numbers come in lower than a vendor calculator’s, and honest.

The 15 tools

lookup_gcp_billing_skuWhat a mystery Google Cloud line item is, and what triggers it

get_aws_rateA verified AWS list price, with its source and verification date

list_rate_freshnessWhen every catalog was last checked, so answers can be caveated honestly

estimate_nat_gateway_costNAT hours, processing, egress and cross-AZ — the charges that stack

compare_ebs_gp2_gp3Including the paid IOPS that can make the migration a loss

compare_s3_storage_classesWith transitions, the 128 KB floor and minimum durations charged

tune_lambda_memoryThe cheapest setting, and the fastest one that costs the same

compare_fargate_vs_ec2Charging EC2 for whole instances, packing loss and idle hours

compare_dynamodb_capacity_modesWhere the on-demand break-even actually sits

compare_cloudfront_vs_s3Decided by average object size, not cache hit rate

estimate_kubernetes_costSized from pod requests, with the non-compute charges added

estimate_ai_agent_costModelling the loop, where cost grows with the square of step count

list_llm_modelsModels with verified token prices

explain_bigquery_cost_anti_patternWhy a piece of SQL costs what it does

find_finops_sqlA copy-paste INFORMATION_SCHEMA query for a cost question

Freshness

Rates are compiled in at publish time and each carries a verification date. The oldest AWS rate in the current build was read on 4 Aug 2026, across 11 rate groups. The publishing repo runs a build gate that fails on anything past its limit, so a stale package is one that has not been republished rather than one with quietly wrong numbers — and list_rate_freshness lets an assistant check before it answers.

This one answers from a rate card. The product answers from your bill.

Everything here is list prices and models. Finitizer connects read-only and keyless to your actual accounts, applies your negotiated rates, and finds the specific resources costing money — then ships the fix as a Terraform PR or a Jira ticket. There is an authenticated MCP server for that too.

Get a free savings analysis See how Finitizer works