MiniMax-MCP

Converts text input to audio using MiniMax's text-to-audio API through a FastMCP server decorator pattern. The implementation exposes a @mcp.tool decorated function that accepts text and voice parameters, validates inputs, routes requests through the MiniMax API client, and returns either direct URLs (url mode) or downloads audio files locally (local mode) based on MINIMAX_API_RESOURCE_MODE configuration. Supports regional API endpoints (global vs mainland China) with region-specific API keys.

Exposes a list_voices MCP tool that queries MiniMax's voice catalog and returns available voice identifiers and metadata. The implementation calls the MiniMax API client's voice listing endpoint, caches results in memory during server runtime, and returns structured voice data (voice IDs, names, language support, characteristics) to enable client-side voice selection UI or programmatic voice filtering. Supports both global and region-specific voice catalogs.

Uses the FastMCP framework to register MiniMax capabilities as discoverable MCP tools with standardized JSON schemas. Each tool is decorated with @mcp.tool and includes parameter definitions, descriptions, and return types that FastMCP automatically exposes to MCP clients. The framework handles schema generation, parameter validation, and error serialization according to MCP specification. Clients can introspect available tools and their schemas without hardcoding tool knowledge.

Implements error handling for MiniMax API failures, network timeouts, and invalid parameters. The server catches API exceptions, validates inputs before invocation, and returns structured error messages to clients according to MCP error specification. Likely includes retry logic for transient failures (network timeouts) and graceful degradation for permanent failures (invalid API keys, quota exceeded). Error messages include diagnostic information to aid debugging.

Implements a voice_clone MCP tool that accepts one or more audio file samples and generates a cloned voice profile in MiniMax's voice synthesis system. The tool handles audio file upload/streaming to the MiniMax API, manages the voice cloning training process (which may be asynchronous), and returns a voice_id for the cloned voice that can be used with text_to_audio. Supports both local file paths and URL-based audio sources depending on client capabilities.

Implements a generate_video MCP tool that accepts a text prompt and optional generation parameters (duration, resolution, style, etc.) and invokes MiniMax's video generation API. The tool handles prompt validation, parameter marshaling to MiniMax API format, manages the asynchronous video generation process, and returns video URLs or local file paths based on resource mode configuration. Supports polling for generation status and handles long-running generation jobs transparently.

Implements a text_to_image MCP tool that accepts a text prompt and optional generation parameters (style, resolution, aspect ratio, etc.) and invokes MiniMax's image generation API. The tool validates prompts, marshals parameters to API format, handles the image generation process, and returns image URLs or local file paths based on MINIMAX_API_RESOURCE_MODE configuration. Supports batch image generation and style/quality parameters for fine-grained control.

Implements a play_audio MCP tool that plays audio files on the local system where the MCP server is running. The tool accepts a file path (local filesystem path or URL) and invokes the system audio player (likely using Python's subprocess or platform-specific audio libraries). Enables real-time audio preview during development or testing without requiring external audio player applications.

Implements a transport layer that abstracts communication between MCP clients and the MiniMax MCP server using two protocols: stdio (standard input/output for local execution) and SSE (Server-Sent Events for network-based deployment). The FastMCP framework handles protocol negotiation, message serialization/deserialization, and connection lifecycle management. Clients can choose transport based on deployment model (local vs cloud) without changing tool implementations.

Implements a resource handling abstraction controlled by MINIMAX_API_RESOURCE_MODE environment variable that determines how generated media (audio, video, images) is delivered to clients. In 'url' mode (default), the server returns direct URLs to MiniMax CDN resources. In 'local' mode, the server downloads resources to the local filesystem at MINIMAX_MCP_BASE_PATH and returns local file paths. This abstraction allows clients to choose storage strategy (cloud URLs vs local files) without tool implementation changes.

Implements API endpoint routing logic that selects between global and mainland China MiniMax API servers based on configuration. The server reads region configuration (likely from environment variables or config files) and routes all API requests to the appropriate endpoint: https://api.minimaxi.chat (global) or https://api.minimax.chat (mainland China). API keys are region-specific and must match the selected endpoint. This abstraction enables single codebase deployment to multiple regions without code changes.

Provides configuration templates and integration guidance for connecting the MiniMax MCP server to popular MCP-compatible client applications: Claude Desktop, Cursor, and Windsurf. Integration involves adding server configuration (command, arguments, environment variables) to client configuration files (claude_desktop_config.json for Claude Desktop, cursor settings for Cursor, etc.). Once configured, clients automatically discover and invoke all MiniMax tools (text-to-audio, voice-clone, generate-video, text-to-image) without additional setup.