MiniMax CLI
The official mmx CLI of the MiniMax platform with an agent skill: text, images, video, speech, audio transcription, vision and search
- 2,209 ★ on the repository
- Last commit
Extended access
Why this level
- Commands send your prompts, texts, images and audio to MiniMax servers and spend plan balance
- The API key or OAuth tokens are stored in files under `~/.mmx`
- `mmx agent setup` changes the configuration files of chosen agents and in interactive mode can install them
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/minimax-cliDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Commands send your prompts, texts, images and audio to MiniMax servers and spend plan balance
- The API key or OAuth tokens are stored in files under `~/.mmx`
- `mmx agent setup` changes the configuration files of chosen agents and in interactive mode can install them
This is third-party code. Review the repository files before installing.
What it does
MiniMax CLI installs the `mmx` command, through which an agent or a person reaches MiniMax platform models straight from the terminal. `mmx text chat` runs a dialogue with a model with streaming output, a system prompt, history from a file and tool definitions, `mmx image` creates images from a description with aspect ratio control and batches, `mmx video generate` starts asynchronous video generation (the Hailuo-2.3 and MiniMax-H3 models, the latter with a start frame, reference images and video) and shows progress. `mmx speech synthesize` voices text with a choice of voices and speed control, `mmx speech transcribe` transcribes audio to json, srt or vtt subtitles, `mmx vision` describes an image, and `mmx search` searches the web with MiniMax. For agents the repository carries the `mmx-cli` skill describing every command and flags for non-interactive use: `--non-interactive`, `--quiet`, `--output json`, `--async`, `--dry-run` and `--yes`. A separate `h3-video` skill covers the MiniMax-H3 video model. `mmx agent setup` configures MiniMax as the model provider for Claude Code, Codex, Grok Build, OpenCode, Hermes, Pi and Oh my Pi, verifies the key and region before writing and makes a backup before changing files. Authorization uses OAuth with a device code and PKCE or an API key: `mmx auth login`, `mmx auth login --api-key`. The Global (`api.minimax.io`) and CN (`api.minimaxi.com`) regions are detected automatically, and `mmx quota` shows the remaining limit.
Who it is for. For developers and content creators who have a MiniMax plan or key and want text, image, video and speech generation straight from the terminal and an agent.
Good fit when
- An agent needs to generate an image, a clip or a voice-over with a command in a script, with json output to parse
- You need to transcribe a meeting or call recording into text or subtitles, including a mode with speakers and timestamps
- You want to quickly ask a MiniMax model or search the web with the platform without a separate server
- You want to connect MiniMax as a model provider in several agents at once with one `mmx agent setup` command
- Tasks run in CI or a service: the key is passed with `--api-key`, and the settings folder is set by `MMX_CONFIG_DIR`
Not a fit when
- You have no MiniMax access: the CLI talks only to the MiniMax platform
- You need models from other providers: the `mmx` commands call only MiniMax models
- Audio is longer than 500 seconds or larger than 50 MB: such files are not accepted for transcription
- Search must return more than 10 results per query: the search interface has no pagination
Example request
Generate a picture of a cat in a spacesuit with mmx and save it to the out folderLimitations
Node.js 18 or newer and access to MiniMax are required: the README asks for a Token Plan, while agent setup distinguishes a subscription from pay-as-you-go `sk-api-...` keys. The preview model `MiniMax-M3.1-Flash-Preview`, which the CLI picks by default, is currently available on the M Plan or MiniMax Code; a pay-as-you-go key needs an explicit `--model MiniMax-M3`. Transcription accepts wav, aiff, flac, m4a, mp3, aac, opus and ogg up to 50 MB and 500 seconds, and a file above 50 MB is rejected locally. For MiniMax-H3 sizes are checked up front: an image up to 30 MB, a reference video up to 50 MB, audio up to 15 MB, the whole request up to 64 MB.
Search returns up to 10 results per call. Data goes to MiniMax servers in the chosen region. The README states the MIT license, but there is no LICENSE file in the repository and GitHub detects no license: rights to use and copy are not defined, check with the author. The repository belongs to MiniMax itself, and the version in `package.json` is `0.0.0-dev`, the real one is derived from the tag at release.
How to disable. Remove the package with `npm uninstall -g mmx-cli` and the skill from the agent's skill manager. Sign out with `mmx auth logout`; settings and tokens live in `~/.mmx` (or in the folder from `MMX_CONFIG_DIR`). Restore agent configurations written by `mmx agent setup` from the backups it created.
FAQ
How do I sign in?
`mmx auth login` offers OAuth (device code per RFC 8628 with PKCE, the browser opens for entering the code) or pasting an API key. For CI use `mmx auth login --api-key sk-xxxxx`. OAuth and a key are mutually exclusive: logging in one way clears the other. The `--api-key` flag has priority, then OAuth, then the key from settings.
How does video generation work?
`mmx video generate --prompt ...` starts a task and waits for the result by default; with `--async` it returns an identifier at once. `mmx video task get --task-id` shows status and `mmx video download --file-id` fetches the file. For MiniMax-H3 text-to-video defaults to 2K, 5 seconds and 16:9.
What does `mmx agent setup` do?
It configures MiniMax for Claude Code, Codex, Grok Build, OpenCode, Hermes, Pi and Oh my Pi while keeping unrelated settings. By default it picks `MiniMax-M3.1-Flash-Preview` for a subscription and `MiniMax-M3` for pay-as-you-go `sk-api-...` keys. `--dry-run` shows changes without a live request, and the command never launches an agent.
Which regions and addresses are used?
Global at `api.minimax.io` and CN at `api.minimaxi.com`. The region is detected automatically (both are probed for a key) and set with the `--region` flag or the `MINIMAX_REGION` variable. The address can be overridden with `MINIMAX_BASE_URL`.
Can the CLI be used as a library?
Yes. The package exports `mmx-cli/sdk` with TypeScript types, described in `SDK.md`. Scripts also have environment variables: `MINIMAX_OUTPUT`, `MINIMAX_TIMEOUT` and `MINIMAX_VERBOSE` for a detailed request log.
Discussion
No questions or reviews yet.
Related
CLI toolsOfficial MiniMax MCP server: text to speech, voice cloning and design, image and video generation
An MCP server for Autodesk 3ds Max: the agent reads the scene, models, and sets up materials and modifiers
Local speech synthesis for agents: a command and a skill voice text, clone a voice from a recording and render scores to audio, with no cloud
A set of skills for agents to generate and edit images, video and audio through the muapi.ai service and muapi-cli
Higgsfield CLI
higgsfield-cli
The official Higgsfield command-line tool: more than 40 models for images, video, 3D and audio and analysis of finished video from the terminal
A skill that writes prompts for video, image and music generation on 14 platforms and can submit them in the browser
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| MiniMax CLI (this entry) | yes | 2,209 | ||
| MiniMax MCP | yes | API key | 1,594 | the official MiniMax MCP server: speech synthesis, voice cloning, image and video generation over MCP |
| MiniMax Skills | yes | 13,742 | 17 MiniMax skills for development and documents, without terminal access to generation models |
README in short
The description in English and Chinese calls the project the official CLI of the platform and lists features: text, images, video, speech, vision, search and two regions. Installation, a quick start, detailed sections per command group with examples, environment variables, agent setup and updating follow. The repository holds an SDK reference, a description of errors, notes on the CLI design and releases, and two agent skills.