Speech, images, video and music sit behind one connection, and each generator writes into a directory you pass rather than handing back a URL to collect later. Long renders do not have to block: video generation can run asynchronously and return a task you poll. The costs are real per call, so the tools are worth wiring where a person asks for an asset rather than where a model might decide one would help.
MiniMax's generation APIs behind ten tools: text to speech with voice, emotion and speed control, voice cloning and voice design, text to image, text or image to video, music generation, and local playback.
- Text spoken in a chosen voice with emotion, speed, pitch, volume, format, bitrate and sample rate, saved to a directory you name
- Subtitles produced alongside the audio when you ask for them
- The available voices listed, when the account is on the host that supports it
- A voice cloned from an audio file you supply, and a new voice designed from a written description, with preview audio
- Images generated from a prompt with aspect ratio, a count, and an optional prompt optimizer
- Video generated from a prompt or from a first-frame image, with duration and resolution, and an async mode whose task you poll by id
- Music generated from a prompt for the style and the lyrics to sing
- WAV and MP3 files played back locally
A MiniMax account with an API key in MINIMAX_API_KEY and the matching host in MINIMAX_API_HOST — the international and mainland platforms issue different keys, and a key from one will not authenticate against the other. MINIMAX_MCP_BASE_PATH sets where generated files are written and MINIMAX_RESOURCE_MODE whether they come back as files or URLs. Node with npx. Every generation tool spends credits on the account, which is why their descriptions ask to be called only when the user actually asked for output. MIT.
One command plus a key — npx -y minimax-mcp-js, then supply credentials