Show HN: Audio Toolkit for Agents

Article URL: https://github.com/shiehn/sas-audio-processor

Comments URL: https://news.ycombinator.com/item?id=47207806

Points: 32

Comments: 4

SAS Audio Processor

This is a component of Signals & Sorcery, a music production application.

A suite of 25 audio processing tools — trim, normalize, compress, EQ, reverb, pitch-shift, time-stretch, key detection, MIDI extraction, and more — exposed as MCP tools via DeclarAgent.

What You Get

********************

Every tool accepts WAV files and outputs structured JSON.

Quick Start (Claude Code)

This is the primary use case — use these audio tools directly from Claude Code via MCP.

1. Install DeclarAgent

go install github.com/stevehiehn/declaragent@latest

2. Install sas-audio-processor

# From source
git clone https://github.com/stevehiehn/sas-audio-processor.git
cd sas-audio-processor
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt && pip install -e .

Or use the pre-built binary (macOS only):

./build.sh
# Binary at dist/sas-processor/sas-processor

3. Add MCP config

Add to ~/.claude/settings.json:

"mcpServers": {
"sas-audio": {
"command": "declaragent",
"args": ["--plans-dir", "/path/to/sas-audio-processor/plans"]

4. Use it

Now Claude Code has access to all 25 audio tools. Example conversation:

You: Analyze this drum loop and tell me the BPM
Claude: [calls analyze tool] The drum loop is at 128 BPM, 44100 Hz, stereo, 8.2 seconds long.

You: Trim it to 4 bars and normalize to -14 LUFS
Claude: [calls trim tool, then normalize tool] Done — trimmed to 4 bars starting at the downbeat, normalized to -14 LUFS.

You: Now master it
Claude: [calls master-track tool] Applied normalize → compress → limit chain. Output saved.

Quick Start (Other MCP Clients)

Claude Desktop

Add to claude_desktop_config.json:

"mcpServers": {
"sas-audio": {
"command": "declaragent",
"args": ["--plans-dir", "/path/to/sas-audio-processor/plans"]

Cursor / Windsurf / Copilot

Add the same MCP server config to your editor's MCP settings. The command and args are identical — only the config file location differs per editor.

How It Works

MCP Client (Claude Code, Claude Desktop, etc.)
│
└─► DeclarAgent (reads YAML plan files from plans/)
│
└─► sas-processor <subcommand> --input ... --output ...
│
└─► stdout: JSON result

Each YAML plan file in plans/ defines one MCP tool. DeclarAgent reads these at startup and exposes them to MCP clients. When a tool is called, DeclarAgent runs the sas-processor CLI under the hood.

Available Tools

Processing

| Tool | Description |
| --- | --- |
| trim | Detect downbeat and trim audio to specified number of bars |
| time-stretch | Change tempo while preserving pitch |
| convert | Change sample rate and/or bit depth |
| mono | Convert stereo to mono |
| silence-remove | Trim silence from start and end |
| split-bars | Split audio into N-bar chunks |

#### Effects

Analysis

| Tool | Description |
| --- | --- |
| analyze | Detect BPM, duration, sample rate, channels |
| detect-key | Detect musical key and mode (major/minor) |
| loudness | Measure integrated LUFS and peak dB |
| onset-detect | Detect onset/transient times |

#### MIDI

| Tool | Description |
| --- | --- |
| melody-to-midi | Extract monophonic melody to MIDI file |

#### Composite (Multi-Step)

CLI Reference

All subcommands output line-delimited JSON to stdout. Errors go to stderr as JSON.

sas-processor <subcommand> [options]

ping

sas-processor ping

Health check — returns version and status.

analyze

sas-processor analyze --input <file>

Returns BPM, duration, sample rate, and channel count.

trim

sas-processor trim --input <file> --output <file> --bpm <float> --bars <int> [--meter 4] [--verbose]

Detects the downbeat and trims audio to the specified number of bars.

time-stretch

sas-processor time-stretch --input <file> --output <file> --source-bpm <float> --target-bpm <float>

Changes tempo while preserving pitch.

normalize

sas-processor normalize --input <file> --output <file> [--mode lufs|peak] [--target-lufs -14] [--target-peak -1]

gain

sas-processor gain --input <file> --output <file> --db <float>

mono

sas-processor mono --input <file> --output <file>

convert

sas-processor convert --input <file> --output <file> [--sample-rate <int>] [--bit-depth 16|24|32]

silence-remove

sas-processor silence-remove --input <file> --output <file> [--top-db 30]

compress

sas-processor compress --input <file> --output <file> [--threshold -20] [--ratio 4] [--attack 1.0] [--release 100]

eq

sas-processor eq --input <file> --output <file> --freq <float> --gain-db <float> [--q 1.0]

reverb

sas-processor reverb --input <file> --output <file> [--room-size 0.5] [--damping 0.5] [--wet-level 0.33]

limit

sas-processor limit --input <file> --output <file> [--threshold -1]

filter

sas-processor filter --input <file> --output <file> --type highpass|lowpass --cutoff-hz <float>

pitch-shift

sas-processor pitch-shift --input <file> --output <file> --semitones <float>

detect-key

sas-processor detect-key --input <file>

Returns key, mode (major/minor), and confidence score.

loudness

sas-processor loudness --input <file>

Returns integrated LUFS and peak dB.

onset-detect

sas-processor onset-detect --input <file>

Returns onset count and onset times in seconds.

split-bars

sas-processor split-bars --input <file> --output-dir <dir> --bpm <float> [--bars-per-chunk 1] [--meter 4]

melody-to-midi

sas-processor melody-to-midi --input <file> --output <file.mid>

Building from Source

Prerequisites

• macOS 10.15+

• Python 3.9+

Build Steps

git clone https://github.com/stevehiehn/sas-audio-processor.git
cd sas-audio-processor
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt && pip install -e .

# Build self-contained binary
pyinstaller sas-processor.spec --clean --noconfirm
# Output: dist/sas-processor/

Or use the build script:

./build.sh

Cross-Architecture Builds

Apple Silicon (ARM64):

pyinstaller sas-processor.spec --clean --noconfirm
mv dist/sas-processor dist/sas-processor-arm64

Intel (x86_64):

arch -x86_64 /bin/bash -c "
python3 -m venv venv-x64
source venv-x64/bin/activate
pip install -r requirements.txt
pyinstaller sas-processor.spec --clean --noconfirm
mv dist/sas-processor dist/sas-processor-x86_64

Running Tests

source venv/bin/activate

# All tests
pytest tests/ -v

[...]

---

*[Original source](https://github.com/shiehn/sas-audio-processor)*