Summary

Speech, a cloned voice, an image or a video, produced from inside the conversation.

Video generation is asynchronous — the generate call starts the job and a separate query call collects it — so it is worth telling your client the completion rule up front rather than having it wait on the first response.

What it is

The official MiniMax MCP server, connecting an MCP client to MiniMax's text-to-speech and image and video generation APIs. It runs over stdio locally, or SSE when deployed to a server.

What you get

`text_to_audio` — convert text to audio with a given voice
`list_voices` — list every voice available to your account
`voice_clone` — clone a voice from audio files you supply
`voice_design` — generate a voice from a descriptive prompt, with preview text
`text_to_image` — generate an image from a prompt
`generate_video` — generate video from a prompt, including the MiniMax-Hailuo-02 model with 6s and 10s durations and 768P or 1080P output
`query_video_generation` — poll a video generation task for its result, which is how the asynchronous path is driven

Requirements

`uv` installed, then run it with `uvx minimax-mcp`. Two environment variables must agree: `MINIMAX_API_KEY` and `MINIMAX_API_HOST`, where the global host is `https://api.minimax.io` and the mainland host is `https://api.minimaxi.com`. A key from one region against the other's host returns an invalid-API-key error, which is the most common setup failure. `MINIMAX_MCP_BASE_PATH` sets where generated files land locally, and `MINIMAX_API_RESOURCE_MODE` chooses between a returned URL and a local download. On Windows, Claude Desktop needs Developer Mode enabled. Calls to these tools consume MiniMax credits.

Setup effort

One command plus a key — uvx minimax-mcp -y, then supply credentials