MiniMax MCP is a Model Context Protocol (MCP) server that exposes MiniMax's generation APIs to MCP clients. Once configured with an API key, a client can synthesise speech, list and design voices, clone a voice from sample audio, generate images, and start and poll video generation jobs.
The server runs over either of two transports: stdio for local use, communicating on standard output, and Server-Sent Events for a local or cloud deployment, communicating over the network. Local file paths or URLs are accepted as input for the stdio transport; a cloud deployment takes URLs. Generated audio, image and video results are returned either as URLs or downloaded to a local output directory, controlled by a resource-mode setting.
The API key and API host are region-paired: a global key works with the global host and a mainland key with the mainland host, and a mismatch is reported as an invalid key. A separate JavaScript implementation of the same server exists, and MiniMax also publishes a command-line tool covering text, vision and search.
Features
- Text to speech: texttoaudio converts text to audio with a chosen voice, and list_voices enumerates what is available
- Voice cloning: voice_clone builds a voice from supplied audio files
- Voice design: voice_design generates a voice from a descriptive prompt and returns preview audio
- Image generation: texttoimage produces an image from a prompt
- Video generation: generatevideo starts a job and queryvideo_generation polls its result, with the MiniMax-Hailuo-02 model offering 6s or 10s durations at 768P or 1080P
- Two transports: stdio for local runs, SSE for local or cloud deployment
- Resource mode: results delivered as URLs or written to a local output directory
- Region pairing: separate global and mainland API hosts, each matched to its own key
