MiniMax MCP

Seven tools cover the generative surface: speech synthesis with a listed voice, voice cloning from sample audio, voice design from a text prompt, image generation, and video generation. Video is deliberately split in two, because a Hailuo render takes long enough that holding a tool call open is the wrong shape. The generation tool returns a task identifier and a separate query tool polls it, so the agent keeps working while the render proceeds. One configuration trap is worth knowing before the first call: the global and mainland deployments use different API hosts and separate keys, and a key from one region against the other region's host simply fails authentication.