← Back to Skills Marketplace
2655
Downloads
0
Stars
8
Active Installs
8
Versions
Install in OpenClaw
/install mlx-audio-server
Description
Local 24x7 OpenAI-compatible API server for STT/TTS, powered by MLX on your Mac.
Usage Guidance
This skill appears to do what it says (install a local MLX audio server and provide STT/TTS helpers), but take these precautions before installing:
- Review the Homebrew tap/formula (guoqiao/tap) referenced by install.sh on GitHub to ensure the code and install steps are trustworthy. Third-party taps can install arbitrary software.
- Prefer running install.sh manually yourself (inspect it line-by-line) rather than letting an agent run it autonomously. The script updates brew, installs packages, and starts a LaunchAgent service.
- Because the skill is marked always: true, it will be force-included in every agent run; consider removing or disabling the skill unless you need that behavior.
- No credentials are requested by the skill, and the runtime scripts only contact localhost for the API, but network activity to fetch the Homebrew tap and packages will occur. Only install on a machine you control and trust.
If you want help auditing the Homebrew formula or reviewing the tap's repository, provide the tap URL and I can list the specific install actions to check.
Capability Analysis
Type: OpenClaw Skill
Name: mlx-audio-server
Version: 0.2.2
The skill is classified as suspicious primarily due to its reliance on an external Homebrew formula from a custom tap (`guoqiao/tap/mlx-audio-server`) in `install.sh`. This introduces a supply chain risk, as the content of this external formula is not part of the analyzed bundle and could potentially be altered or malicious. Additionally, the `install.sh` script establishes persistence by setting up `mlx-audio-server` as a LaunchAgent service on macOS, a high-risk capability, even though it aligns with the stated purpose of running a local 24x7 API server. While other scripts (`run_stt.sh`, `run_tts.sh`) only make local network calls to `localhost:8899`, the external dependency and persistence mechanism elevate the risk beyond benign.
Capability Assessment
Purpose & Capability
Name and description match the provided scripts and README: the skill installs a Homebrew formula for an MLX audio server, exposes OpenAI-compatible endpoints on localhost, and includes helper scripts for STT/TTS. Required binary (brew) and ffmpeg/jq usage are appropriate for the stated purpose.
Instruction Scope
SKILL.md and the scripts limit actions to installing the Homebrew formula, starting/restarting a brew service, converting audio with ffmpeg, and POSTing to localhost:8899. They do not read unrelated system files or exfiltrate data to remote endpoints beyond standard Homebrew/GitHub network operations.
Install Mechanism
There is no formal install spec; the provided install.sh runs 'brew install guoqiao/tap/mlx-audio-server' and 'brew services restart'. This pulls code from a third-party Homebrew tap (guoqiao). Installing a third-party tap via brew is a moderate risk because it executes/installs upstream code not bundled with the skill and the tap/source should be reviewed before trust. No direct download URLs are embedded in the skill, but the tap may fetch arbitrary code.
Credentials
The skill requests no secrets or credentials. It optionally reads MLX_AUDIO_SERVER_PORT for the local port. There are no requested env vars that don't match the task.
Persistence & Privilege
The skill is marked always: true, which force-includes it in every agent run. Combined with the install script that registers a LaunchAgent (brew services) and restarts a service, this grants the skill ongoing system presence without clear justification. While running a local service fits the skill's purpose, always: true is broader than necessary and increases blast radius if the tapped formula or service behaved unexpectedly.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install mlx-audio-server - After installation, invoke the skill by name or use
/mlx-audio-server - Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.2.2
- Updated description to highlight always-on, local API server capabilities for STT/TTS on Mac.
- Simplified and clarified project overview in SKILL.md.
- Removed the run_stt.py script from the project.
v0.2.1
- Updated SKILL.md to clarify the Homebrew tap source, linking directly to the guoqiao/tap repository.
- Documented default models used for STT (mlx-community/glm-asr-nano-2512-8bit) and TTS (mlx-community/Qwen3-TTS-12Hz-1.7B-VoiceDesign-bf16).
- Improved installation instructions to specify the tap used for the Homebrew formula.
v0.2.0
mlx-audio-server 0.2.0
- Switched installation to use a Homebrew formula (`mlx-audio-server`) for easier setup and management.
- Now installs required dependencies (`ffmpeg`, `jq`) via Homebrew automatically.
- Added new files: `README.md` and `run_stt.py`.
- Updated documentation to reflect the improved installation process and usage instructions.
v0.1.4
- No file changes detected in this release.
- Version number updated to 0.1.4.
v0.1.3
No changes detected in this version.
v0.1.2
- Improved description for clarity: highlights OpenAI compatibility, local operation, and Apple Silicon support.
- Clarified that the server runs in the background as a launchd service.
- Updated default model names and descriptions for both STT and TTS.
- Added note on initial model downloads for first run performance.
- Made instructions and usage examples more concise and user-friendly.
v0.1.1
- Improved clarity in model names and usage instructions.
- Updated default model names to be inline code blocks for emphasis.
- Clarified output formats for both STT and TTS usage.
- Minor wording and formatting enhancements for better readability.
v0.1.0
mlx-audio-server 0.1.0
- Initial release of an audio processing server based on Apple's MLX framework for Apple Silicon.
- Provides fast and efficient text-to-speech (TTS), speech-to-text (STT), and speech-to-speech (STS) functionality.
- Runs as an OpenAI-compatible API server by default on port 8899.
- Includes default models for STT (glm-asr-nano-2512-8bit) and TTS (Qwen3-TTS-12Hz-1.7B-VoiceDesign-bf16).
- Offers simple install and usage scripts, leveraging brew and uv for environment setup.
- Supports macOS with Apple Silicon only.
Metadata
Frequently Asked Questions
What is MLX Audio Server?
Local 24x7 OpenAI-compatible API server for STT/TTS, powered by MLX on your Mac. It is an AI Agent Skill for Claude Code / OpenClaw, with 2655 downloads so far.
How do I install MLX Audio Server?
Run "/install mlx-audio-server" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is MLX Audio Server free?
Yes, MLX Audio Server is completely free (open-source). You can download, install and use it at no cost.
Which platforms does MLX Audio Server support?
MLX Audio Server is cross-platform and runs anywhere OpenClaw / Claude Code is available (darwin).
Who created MLX Audio Server?
It is built and maintained by guoqiao (@guoqiao); the current version is v0.2.2.
More Skills