Overview
What is Luma AI MCP Server?
An MCP server that integrates with Luma AI's Dream Machine API (v1) to let AI assistants generate, manage, and manipulate AI-generated videos and images. It is intended for developers and users who want to control Luma's creative tools through LLMs like Claude.
How to use Luma AI MCP Server?
Obtain a Luma API key from Luma AI, then add the server configuration to your Claude Desktop configuration file (~/Library/Application Support/Claude/claude_desktop_config.json on macOS). Use uv run --project /path/to/server -m luma_ai_mcp_server as the command, and set the LUMA_API_KEY environment variable. After restarting Claude Desktop, the tools become available.
Key features of Luma AI MCP Server
- Text-to-video generation
- Advanced video generation with keyframes
- Image-to-video conversion
- Video extension and interpolation
- Image generation with reference images
- Audio addition to videos
- Video upscaling
- Credit management and generation status tracking
Use cases of Luma AI MCP Server
- Create short AI-generated videos from text prompts
- Extend or interpolate between existing videos using keyframes
- Upscale a completed video to a higher resolution
- Add AI-generated audio to a video based on a description
- Generate images with character or style references
FAQ from Luma AI MCP Server
What does the Luma AI MCP Server do?
It provides a set of tools that allow an AI assistant to call Luma AI's Dream Machine API to generate, check status, upscale, add audio to, and delete video generations, as well as generate images and manage credits.
What are the runtime dependencies?
The server requires Python with uv and a valid Luma API key set as the LUMA_API_KEY environment variable. It communicates over MCP standard input/output (stdio) as configured for Claude Desktop.
Where are generated videos and images stored?
The API returns URLs pointing to the generated content on Luma's servers. The server itself does not store files locally.
What are the limitations of video generation?
Supported durations are only "5s" or "9s". Each completed generation can be upscaled only once, and the target resolution must be higher than the original. Valid resolutions are 540p, 720p, 1080p, and 4k.
How does the server communicate with the client?
It uses the Model Context Protocol (MCP) over standard input/output (stdio). The server is invoked via a command-line tool (uv run) and receives/sends JSON messages.
