Audionaut occupies a gap nobody else has seriously tried to fill: a lightweight multitrack audio editor designed from the ground up to be operated by AI agents. Not a DAW with an API bolted on. Not a CLI wrapper around FFmpeg. A real desktop application with a GUI, per-track playlists, flexible multi-channel support, and clean exports — that also exposes its full editing surface over MCP (Model Context Protocol) so Claude or any compatible agent can drive it programmatically. The technical stack is serious. Written in modern C++ on the JUCE framework, it runs natively on Windows, macOS, and Linux. Audio analysis features — BIC segmentation, onset detection, beat tracking — come from a static build of the Essentia library. Stem separation uses demucs.cpp, a C++ port of Meta's Demucs model, compiled from source with model weights downloaded on first use. The architecture is clean: submodules for dependencies, Catch2 for tests, a standalone CLI tool (audionaut-cli) for headless operation. The MCP integration is the headline feature. You run claude mcp add audionaut -- npx -y audionaut-mcp, keep your project open in the app, and each AI-driven edit arrives as a single undo step. The CLI supports the full workflow: create, import, analyze, auto-edit, assemble, export — all with --json output for machine consumption. Exit codes are well-defined (0 success, 1 failed, 2 usage error, 3 feature unavailable). This is infrastructure-grade thinking about agent interoperability. The CLI command set reveals the real ambition. You can create regions, place and move clips, set clip gain and fades in beats or seconds, run stem separation, split at bar positions, auto-edit by measures, and assemble random arrangements — all headless. Projects written by the CLI open in the GUI and vice versa. The macOS app even accepts CLI verbs directly, executing headlessly while a GUI instance runs. Practical details matter here. The Essentia build requires Python ≤ 3.11 (its bundled waf needs distutils, removed in 3.12). CMake 3.x is recommended because some third-party deps ship ancient CMakeLists that CMake 4.x rejects. Building Essentia is optional — without it, analysis features disable gracefully. The project is dual-licensed under GPL3 and a commercial license, with a CLA for contributors and opt-in anonymous analytics. The honest limitation is that this is early-stage open-source software from what appears to be a solo developer (sponsorship-funded). The feature set is deliberately constrained — no plugin hosting, no mixing console, no MIDI. It's a multitrack editor, not a DAW. For its intended use case — quick audio editing by humans or agents without the overhead of Pro Tools or Reaper — that constraint is the point. What makes Audionaut interesting isn't any single feature. It's the bet that AI agents will need real creative tools they can operate natively, not screen-scrapers or accessibility hacks layered on top of apps designed for mouse clicks. If that bet is right, being first with a clean MCP implementation in a native audio editor is a meaningful position to hold.
