clipwright-noise

MCP tool for noise detection and OTIO timeline annotation generation.

Overview

Measures audio noise floor using ffmpeg astats filter, writes denoise instructions (backend, parameters) to timeline-level metadata["clipwright"]["denoise"].

Performs detection only (OTIO annotation); realization (ffmpeg filter application) is done once by clipwright-render (design M3: separation of detection and application).

Initial render support:

• afftdn backend: render application supported (clipwright-render injects afftdn filter).
• deepfilternet backend: annotation only. render application not yet supported (planned in future version).

Prerequisites

• Python 3.11 or later
• ffmpeg / ffprobe must exist on PATH or full paths set in environment variables CLIPWRIGHT_FFMPEG / CLIPWRIGHT_FFPROBE.

Add ffmpeg to PATH directly or specify via environment variables:

export CLIPWRIGHT_FFMPEG=/path/to/ffmpeg
export CLIPWRIGHT_FFPROBE=/path/to/ffprobe

MCP Tool

clipwright_detect_noise

Parameters

Dependencies

ffmpeg / ffprobe are invoked as separate processes (via PATH or environment variables) for license independence. DeepFilterNet binary is not bundled in initial version; render-side dependency planned.

Installation and Startup

Within a uv workspace:

uv run --package clipwright-noise clipwright-noise

Or install directly:

uv add clipwright-noise
clipwright-noise

MCP Configuration (.mcp.json)

Register clipwright-noise in your MCP host configuration to expose clipwright_detect_noise as a tool.

{
  "mcpServers": {
    "clipwright-noise": {
      "command": "clipwright-noise",
      "env": {
        "CLIPWRIGHT_FFMPEG": "/path/to/ffmpeg",
        "CLIPWRIGHT_FFPROBE": "/path/to/ffprobe"
      }
    }
  }
}

Environment variables:

The console script clipwright-noise (registered in [project.scripts]) launches the MCP server over stdio. No CLI argument parsing is required — all inputs are passed via MCP tool calls.