Luma API MCP

What is Luma API MCP?

Luma API MCP is a Model Context Protocol server that integrates with Luma’s AI generation API to create images and videos from text prompts. It is designed for users of MCP‑compatible clients (e.g., Claude Desktop) who want to generate visual content programmatically with support for image references, style control, and keyframe‑based video.

How to use Luma API MCP?

Install by running setup.sh and pasting your API key (obtain it from https://lumalabs.ai/api/keys). Then open any MCP client, such as Claude Desktop, and invoke the “Create Image” or “Create Video” tools with the parameters described in the README.

Key features of Luma API MCP

• Create images with custom prompts and aspect ratios
• Create videos with configurable duration, resolution, and looping
• Supports multiple models: photon‑1/flash for images, ray‑2/flash/1‑6 for videos
• Reference up to eight images, one style, four character images per generation
• Video keyframe control using images or prior generation IDs
• Typical generation times: images 5–15 s, videos 15–60 s

Use cases of Luma API MCP

• Generate AI images for design concepts and marketing materials
• Produce short AI videos with consistent start/end frames
• Integrate visual content creation into automated MCP workflows
• Rapidly prototype variations using prompt and reference parameters

FAQ from Luma API MCP

Where do I get an API key?

Obtain your API key from https://lumalabs.ai/api/keys.

What MCP clients are supported?

Any MCP client works; the README explicitly mentions Claude Desktop.

What are the default generation settings?

Images default to model photon‑1 and aspect ratio 16:9. Videos default to model ray‑2, resolution 720p, duration 5s, and loop false.

How long does generation take?

Typical image generation takes 5–15 seconds; video generation takes 15–60 seconds depending on duration and resolution.

Can I use reference images?

Yes. For images you can provide up to 8 image references, 1 style reference, up to 4 character references, and 1 image to modify. For videos you can supply start/end frame images or generation IDs.