CloudCostCompareCloud price data in your tools: JSON API and MCP
The numbers on this site are available as plain JSON files, so you can use them in a spreadsheet, a script or a dashboard. There is also a small MCP server, so an AI coding assistant can look up prices while it helps you design a system.
The JSON API
The API is a set of static files under https://cloudcostcompare.net/api/v1/. It is read-only, needs no key and is updated weekly with the rest of the site.
| File | What it holds |
|---|---|
index.json |
The version and every endpoint |
providers.json |
The four clouds, their price source and what the catalog holds for each |
categories.json |
Catalog categories and the files that hold their SKUs |
catalog/<provider>/<category>/<region group>.json |
The SKUs of one shard |
changes.json |
The catalog change log |
ai-models.json |
AI model token prices and published discounts |
examples.json |
Totals for the example workloads |
openapi.json |
An OpenAPI 3.1 description of all of it |
Every response has the same wrapper:
{
"updated_at": "2026-11-03",
"note": "On-demand list prices in USD, read weekly ...",
"data": []
}updated_at is the date the data was read from the vendors, and the note restates what the prices are. Every SKU carries its own source_url and last_verified. A price the vendor does not publish is null, never zero.
Responses are cached for an hour and may be read from a web page in a browser, since they carry a permissive cross-origin header. The files are static, so there is no filtering on the server: fetch a shard and filter it yourself.
A small example
To find the cheapest instances with at least four vCPUs in the US East group, fetch the shard and filter it:
const url = "https://cloudcostcompare.net/api/v1/catalog/aws/virtual-machines/us-east.json";
const { data } = await (await fetch(url)).json();
const fits = data.skus
.filter((s) => (s.attributes.vcpu ?? 0) >= 4)
.sort((a, b) => a.price_usd_per_month_730h - b.price_usd_per_month_730h);Skip any row whose size is missing, rather than treating it as zero.
Please be considerate
There is no stated rate limit. Cache responses, fetch only what you need and do not poll more often than the data changes, which is weekly.
The MCP server
MCP, the Model Context Protocol, is a standard way for an AI assistant to call tools. The site’s MCP server reads the API and offers four read-only tools:
- find_instances: instances with at least a given vCPU and memory, cheapest first.
- search_ai_models: token prices by tier, vendor and open-weight status.
- recent_price_changes: catalog SKUs added, removed or changed.
- compare_example_workloads: totals for the example factories, never calling an incomplete total the cheapest.
Results include the updated_at date and a note, and each row carries its source. The server does not have a tool for pricing your own workload or for looking up regions, because the site does not have that data yet.
It lives in the project’s packages/mcp-server folder and has not been published to a package registry at the time of writing, so you run it from a checkout. Its README has setup snippets for Claude Code, Claude Desktop, Cursor and VS Code.
What to remember
The data is for planning. It is on-demand list prices, read weekly, not live and not your negotiated rate. An assistant that quotes a price from this data should also say when it was read. For how the numbers get here, see how prices are collected and refreshed. The tables behind the shards are explained in how to use the price catalog, and the token prices in AI model token pricing.