# imagedimensions-mcp

An [MCP](https://modelcontextprotocol.io) server that audits the images on any public web page —
**natural vs. rendered dimensions, oversized-image detection, and format breakdown** — so AI agents
(Claude, Cursor, Windsurf, etc.) can check image performance during development.

Powered by [imagedimensions.com](https://imagedimensions.com). The scan runs server-side, so **no
local browser or Chrome install is required.**

## Install / run

```jsonc
// Claude Desktop / any MCP client config
{
  "mcpServers": {
    "imagedimensions": {
      "command": "npx",
      "args": ["-y", "imagedimensions-mcp"]
    }
  }
}
```

## Tool

### `scan_image_dimensions`

| Param | Type | Description |
| --- | --- | --- |
| `url` | string (required) | Public URL of the page to audit. |
| `oversizedThreshold` | number (optional) | Area-overshoot ratio to flag as oversized. Default `4` (≈2× per dimension). |

Returns a text report plus structured content:

- total / visible image counts and CSS-background count
- modern-format share (WebP + AVIF)
- format breakdown
- the list of oversized images (natural → rendered, overshoot ratio, format, src)
- a link to the full visual report on imagedimensions.com

**Example agent uses:** "Audit the images on https://example.com," "Which images on this page are
oversized and hurting LCP?", "What % of this site's images use modern formats?"

## Why "oversized" matters

An image downloaded much larger than the box it renders into wastes bandwidth and slows Largest
Contentful Paint — the most common image-performance mistake on the web. See
[the writeup](https://imagedimensions.com/blog/right-size-web-images).

## Config

- `IMAGEDIMENSIONS_API_BASE` — override the API base (default `https://imagedimensions.com`).

## License

MIT