---
title: "audio-tools"
description: "Deterministic audio manipulation tools and deep learning stems separation for AI agents."
canonical: https://agentpluginsdirectory.com/plugins/audio-tools
last-updated: 2026-10-21
---

# audio-tools
Deterministic audio manipulation tools and deep learning stems separation for AI agents.
- Slug: audio-tools
- Publisher: G. H. Chinoy
- Repository: https://github.com/ghchinoy/audio-tools-for-agents
- Manifest: plugin.json
- Version: 0.1.0
- License: Apache-2.0
- Category (editorial): creative-media
- Skills: 0
- MCP servers: 1 (audio-tools)
- Stars: 1
- Repository created: 2026-09-04
- Repository last pushed: 2026-09-04
- Publisher type: User
- Listing: https://agentpluginsdirectory.com/plugins/audio-tools
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What audio-tools does, in the publisher's words

> Deterministic audio manipulation and local deep-learning stem separation for AI agents.

audio-tools-for-agents provides local, non-cloud audio processing primitives designed specifically for autonomous agents and the agent economy. Its primary capability is multitrack audio stem separation using Meta's HTDemucs (Hybrid Transformer Demucs) models, packaged across four agent-ready surfaces.

Requirements: uv and ffmpeg.

- CPU-Only Default: Guarantees stability across diverse environments.
- Optional Hardware Acceleration: Pass --device auto to leverage Apple Silicon Metal (mps) or NVIDIA CUDA (cuda).
- Fail-Fast Safety: Checks host RAM before execution. Emits actionable error codes (ERR_LOW_MEMORY, ERR_FILE_NOT_FOUND) on failure.
- separate_stems: Isolates vocal and instrument tracks, returning file locations and telemetry.
- inspect_audio: Retrieves duration, sample rate, channels, format, and file size.
- benchmark_system: Reports host core counts, RAM, and hardware acceleration qualification.

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/ghchinoy/audio-tools-for-agents/HEAD/README.md

## MCP servers

- audio-tools: transport: stdio; command: uv run --project ${PLUGIN_ROOT} python -m audio_tools.mcp_server; env: PYTHONUNBUFFERED

Read from the plugin's own mcp.json. Environment variable names only, never values.