MiniMax Releases MMX-CLI: A Command-Line Interface That Gives AI Agents Native Access to Image, Video, Speech, Music, Vision, and Search

MiniMax, the AI research company behind the MiniMax omni-modal model stack, has released MMX-CLI — Node.js-based command-line interface that exposes the MiniMax AI platform’s full suite of generative capabilities, both to human developers working in a terminal and to AI agents running in tools like Cursor, Claude Code, and OpenCode.

What Problem Is MMX-CLI Solving?

Most large language model (LLM)-based agents today are strong at reading and writing text. They can reason over documents, generate code, and respond to multi-turn instructions. But they have no direct path to generate media — no built-in way to synthesize speech, compose music, render a video, or understand an image without a separate integration layer such as the Model Context Protocol (MCP).

Building those integrations typically requires writing custom API wrappers, configuring server-side tooling, and managing authentication separately from whatever agent framework you are using. MMX-CLI is positioned as an alternative approach: expose all of those capabilities as shell commands that an agent can invoke directly, the same way a developer would from a terminal — with zero MCP glue required.

The Seven Modalities

MMX-CLI wraps MiniMax’s full-modal stack into seven generative command groups — mmx text, mmx image, mmx video, mmx speech, mmx music, mmx vision, and mmx search — plus supporting utilities (mmx auth, mmx config, mmx quota, mmx update).

• The mmx text command supports multi-turn chat, streaming output, system prompts, and JSON output mode. It accepts a --model flag to target specific MiniMax model variants such as MiniMax-M2.7-highspeed, with MiniMax-M2.7 as the default.
• The mmx image command generates images from text prompts with controls for aspect ratio (--aspect-ratio) and batch count (--n). It also supports a --subject-ref parameter for subject reference, which enables character or object consistency across multiple generated images — useful for workflows that require visual continuity.
• The mmx video command uses MiniMax-Hailuo-2.3 as its default model, with MiniMax-Hailuo-2.3-Fast available as an alternative. By default, mmx video generate submits a job and polls synchronously until the video is ready. Passing --async or --no-wait changes this behavior: the command returns a task ID immediately, letting the caller check progress separately via mmx video task get --task-id. The command also supports a --first-frame flag for image-conditioned video generation, where a specific image is used as the opening frame of the output video.
• The mmx speech command exposes text-to-speech (TTS) synthesis with more than 30 available voices, speed control, volume and pitch adjustment, subtitle timing data output via --subtitles, and streaming playback support via pipe to a media player. The default model is speech-2.8-hd, with speech-2.6 and speech-02 as alternatives. Input is capped at 10,000 characters.
• The mmx music command, backed by the music-2.5 model, generates music from a text prompt with fine-grained compositional controls including --vocals (e.g. "warm male baritone"), --genre, --mood, --instruments, --tempo, --bpm, --key, and --structure. The --instrumental flag generates music without vocals. An --aigc-watermark flag is also available for embedding an AI-generated content watermark in the output audio.
• mmx vision handles image understanding via a vision-language model (VLM). It accepts a local file path or remote URL — automatically base64-encoding local files — or a pre-uploaded MiniMax file ID. A --prompt flag lets you ask a specific question about the image; the default prompt is "Describe the image."
• mmx search runs a web search query through MiniMax’s own search infrastructure and returns results in text or JSON format.

Technical Architecture

MMX-CLI is written almost entirely in TypeScript (99.8% TS) with strict mode enabled, and uses Bun as the native runtime for development and testing while distributing to npm for compatibility with Node.js 18+ environments. Configuration schema validation uses Zod, and resolution follows a defined precedence order — CLI flags → environment variables → ~/.mmx/config.json → defaults — making deployment straightforward in containerized or CI environments. Dual-region support is built into the API client layer, routing Global users to api.minimax.io and CN users to api.minimaxi.com, switchable via mmx config set --key region --value cn.

Key Takeaways

• MMX-CLI is MiniMax’s official open command-line interface that gives AI agents native access to seven generative modalities — text, image, video, speech, music, vision, and search — without requiring any MCP integration.
• AI agents running in tools like Cursor, Claude Code, and OpenCode can be set up with two commands and a single natural language instruction, after which the agent learns the full command interface on its own from the bundled SKILL.md documentation.
• The CLI is designed for programmatic and agent use, with dedicated flags for non-interactive execution, a clean stdout/stderr separation for safe piping, structured exit codes for error handling, and a schema export feature that lets agent frameworks register mmx commands as JSON tool definitions.
• For AI devs already building agent-based systems, it lowers the integration barrier significantly by consolidating image, video, speech, music, vision, and search generation into a single, well-documented CLI that agents can learn and operate on their own.

Check out the Repo here. Also, feel free to follow us on Twitter and don’t forget to join our 130k+ ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us

Shobha is a data analyst with a proven track record of developing innovative machine-learning solutions that drive business value.