Set up MiniMax MCP Server
Install the official MiniMax MCP via uv, configure your API key and region, and use speech, image, and video generation from your AI client.
- Skill Road
- Set up MiniMax MCP Server
Published on 18.09.2026
The MiniMax MCP Server brings MiniMax's speech synthesis, voice cloning, and image/video generation into an MCP-capable AI client. This guide walks through obtaining an API key, installation, and the key environment variables.
Get an API key and pick a region
First, create an API key in the developer dashboard of the MiniMax platform. Choosing the right region matters: the global endpoint is https://api.minimax.io, and the mainland China endpoint is https://api.minimaxi.com. According to the provider, the API key and host region must match, or authentication fails with an "Invalid API key" error.
Install uv and start the server
The server requires the Python package manager uv, which can be installed via curl -LsSf https://astral.sh/uv/install.sh | sh. After that, uvx minimax-mcp -y starts the server automatically as needed, with no separate installation step. For Node.js environments, MiniMax also offers a standalone JavaScript variant as a separate repository.
Configure your client
Add the server to your client's configuration file. For Claude Desktop, the entry in claude_desktop_config.json looks like this: command uvx, arguments ["minimax-mcp", "-y"], plus the environment variables MINIMAX_API_KEY (your API key), MINIMAX_API_HOST (your chosen region), MINIMAX_MCP_BASE_PATH (local output directory for generated files), and optionally MINIMAX_API_RESOURCE_MODE (url or local, default url). According to the provider, Windows users must additionally enable Developer Mode in Claude Desktop's settings to allow local processes to run.
First steps with audio, image, and video
Test the connection first with list_voices to see the available voices, then generate a first piece of speech with text_to_audio. For custom voices, use voice_clone to derive a voice from an audio sample, or voice_design to create a voice purely from a text description. Images come from text_to_image, and videos from generate_video based on the MiniMax-Hailuo-02 model; since video generation takes longer, check progress with query_video_generation.
Keep an eye on costs
Since there is no permanently free tier according to the provider, and every generation can incur costs based on MiniMax's pricing structure, it's worth checking the official pricing page before embedding the server into automated or high-frequency workflows.
FAQ
Do I need a paid subscription for the MiniMax MCP Server? The documentation does not describe a permanently free tier; according to the provider, using the tools incurs costs based on MiniMax's pricing structure.
Which host endpoint should I use? Use api.minimax.io for global access and api.minimaxi.com for mainland China accounts. The API key and host must belong to the same region.
Can I clone someone else's voice without consent? It's technically possible, but doing so without the recorded person's consent raises legal and ethical concerns; only use voice_clone with your own recordings or explicit permission.
Why does video generation take longer than a normal reply? Video generation runs asynchronously; generate_video starts the job, and query_video_generation then checks its processing status.
Does the server work without Python? For Node.js environments, MiniMax offers a standalone JavaScript variant as a separate repository; the standard path described here, however, uses uv/uvx.
Frequently asked questions
Do I need a paid subscription for the MiniMax MCP Server?
The documentation does not describe a permanently free tier; according to the provider, using the tools incurs costs based on MiniMax's pricing structure.
Which host endpoint should I use?
Use `api.minimax.io` for global access and `api.minimaxi.com` for mainland China accounts. The API key and host must belong to the same region.
Can I clone someone else's voice without consent?
It's technically possible, but doing so without the recorded person's consent raises legal and ethical concerns; only use `voice_clone` with your own recordings or explicit permission.
Why does video generation take longer than a normal reply?
Video generation runs asynchronously; `generate_video` starts the job, and `query_video_generation` then checks its processing status.
Does the server work without Python?
For Node.js environments, MiniMax offers a standalone JavaScript variant as a separate repository; the standard path described here, however, uses `uv`/`uvx`.