MCP Tools Reference

The MCP server exposes two tools. Both render in real Chrome, both cost 1 credit per successful image, and both return a full-resolution CDN URL plus a downscaled preview image so the model can review the result.

The parameters mirror the REST API parameters, with two differences: there is no webhook_url (MCP calls are synchronous), and an extra format argument lets the model choose png or pdf.

generate-image-from-html

Renders HTML and CSS to a PNG or PDF. Best for content the model designs itself: social cards, OG images, invoices, certificates, receipts, charts and more.

Parameter	Type	Required	Description
`html`	string	Yes	Complete, self-contained HTML to render. Put CSS in an inline `<style>` block.
`css`	string	No	Extra CSS injected into the page.
`width`	integer	No	Viewport width in pixels (1-5000).
`height`	integer	No	Viewport height in pixels (1-5000).
`fullpage`	boolean	No	Capture the full scrollable height instead of the viewport.
`dpi`	integer	No	Device pixel ratio (1-4) for higher-resolution output.
`ms_delay`	integer	No	Milliseconds to wait before capture (e.g. to let fonts or animation settle).
`wait_for_selector`	string	No	Wait until this CSS selector appears before capturing.
`format`	string	No	`png` (default) or `pdf`.

screenshot-url

Captures a screenshot of a live, publicly accessible web page.

Parameter	Type	Required	Description
`url`	string	Yes	The fully-qualified URL to capture (`https://...`).
`css`	string	No	CSS injected into the page before capture (e.g. to hide cookie banners).
`width`	integer	No	Viewport width in pixels (1-5000).
`height`	integer	No	Viewport height in pixels (1-5000).
`fullpage`	boolean	No	Capture the full scrollable height instead of the viewport.
`selector`	string	No	Capture only the element matching this CSS selector.
`dpi`	integer	No	Device pixel ratio (1-4).
`ms_delay`	integer	No	Milliseconds to wait before capture.
`wait_for_selector`	string	No	Wait until this CSS selector appears before capturing.

Output

On success, a tool returns two things to the model:

A text part with the full-resolution image URL on the i.html2img.com CDN and your remaining credit balance.
An image part - a downscaled preview of the render, so the model can see what it produced and iterate. PDFs return the URL only.

The full-resolution file always lives at the returned URL. The inline preview is intentionally small so it fits inside the message-size limits of MCP clients.

The image URL is permanent for the lifetime of a paid account, so the model can hand it straight back to you, drop it into a document, or post it onward.

Credits and errors

Each successful image costs 1 credit, exactly like the REST API.
If you are out of credits, the tool returns a clear message instead of an image, and no credit is spent. Top up on the pricing page.
Render failures (timeouts, unreachable pages, invalid input) come back as readable tool errors so the assistant can adjust and try again.

Reminder: inline your images

The renderer does not fetch remote images. Replace any <img src="https://..."> with a base64 data URI (data:image/png;base64,...). Self-contained HTML with an inline <style> block renders most reliably.