Describe the shot in words — the agent calls generate_image, waits for the result, and shows the image right in the chat. You can pass references for editing (image-to-image).
generate_image is a tool of the Clipia MCP server for generating an image from text or editing by references. The agent (Claude, ChatGPT, Cursor) calls it right in the conversation: you give a prompt and optionally aspect ratio, resolution, and up to four reference images. The tool waits for the result for up to 25 seconds and usually returns the ready image together with a compact preview, so the agent can «see» the result and refine it.
| Parameter | Type | Req. | Description |
|---|---|---|---|
prompt | string | yes | Description of the shot (RU or EN) |
model | string | — | Model slug; defaults to the agent’s pick |
aspect_ratio | string | — | e.g. 1:1, 16:9, 9:16 |
resolution | string | — | 1k / 2k / 4k |
num_images | int (1–4) | — | How many variants to generate |
image_urls | string[] (≤4) | — | HTTPS references → image-to-image mode |
include_preview | bool | — | Return an inline preview (default yes) |
{
"name": "generate_image",
"arguments": {
"prompt": "тёмный неоновый постер кофейни, крупный план чашки, пар",
"aspect_ratio": "9:16",
"resolution": "2k"
}
}{
"request_id": "f1c3...9ab",
"status": "COMPLETED",
"model": "nano-banana-2",
"cost_credits": 1,
"output": {
"images": [{
"url": "https://media.clipia.ai/.../preview.webp",
"original_url": "https://media.clipia.ai/.../original.png"
}]
}
}Pass image_urls and the same tool works in image-to-image mode: background swap, styling, edits.
The agent gets a compact preview and can assess the shot and request an edit at once — a vision loop with no tab switching.
With a clipia_test_* key the tool returns an instant mock — debug the integration without spending credits.
claude mcp add --transport http clipia https://api.clipia.ai/mcp \
--header "Authorization: Bearer clipia_live_XXXX"For claude.ai and ChatGPT no key is needed — sign in with your account. Create a key in Clipia settings.
The credit cost is returned in every response (cost_credits) — no hidden MCP markup, the same balance as on the site.
Yes, prompt accepts Russian and English text.
Pass up to 4 HTTPS links in image_urls — image-to-image mode turns on.
The credit cost is in every response (cost_credits) and depends on the model and resolution.