Luma API MCP

What it does

Luma Labs' first-party MCP server, published from the lumalabs GitHub org, exposing exactly two tools: create_image on photon-1 and photon-flash-1, and create_video on ray-2, ray-flash-2 and ray-1-6. create_video takes frame0 and frame1 references, so an agent can pin both ends of a shot and chain continuity across a sequence; create_image accepts image, style, character and modify references for consistent looks. Narrow by design — no polling, extension or upscaling tools — but it is the only Luma-maintained MCP that exists.

Connect

Connection is set up through the vendor's flow — follow the official connect instructions.

Local install: Clone github.com/lumalabs/luma-api-mcp and run `sh setup.sh`, which installs jq, uv and Python 3.10, prompts for LUMA_API_KEY, and registers the server in claude_desktop_config.json as `uv run --python python3.10 --with mcp[cli] --with aiohttp mcp run` against server.py.

Sample tools

Field notes

Official but frozen: MIT, 26 stars, 11 forks, created 2025-04-15 and last pushed 2025-04-18 (still true on 2026-10-21), two open issues, no releases and no package on PyPI. It speaks the Dream Machine API only — Luma's current Agents API (uni-1, uni-1-max, ray-3.2 at agents.lumalabs.ai/v1 under LUMA_AGENTS_API_KEY) is not supported here or in any other MCP server, and the Agents API docs still mention no MCP as of 2026-10-21. The documented install path only writes Claude Desktop's config file; there is no vendor-published `claude mcp add` form, hence the null above rather than a command we made up. For a wider Dream Machine tool surface use the community `luma` server; for ray-3.2 today, reach it through the fal, Replicate or Runway hub MCPs.

Also in Video Generation

Pairs well with