With 144 tools, the pipelines are what make it usable rather than overwhelming: auto_cleanup_podcast or auto_audiobook_mastering run the whole chain — noise, dynamics, loudness — in one call, and check_pipeline_status lets you watch it work. It drives your own local Audacity through the scripting pipe, so the audio and the transcription both stay on your machine; the tradeoff is that Audacity has to be open with mod-script-pipe enabled, which is the usual first-run snag.
A server that gives an AI assistant control of Audacity through its named-pipe scripting interface: 144 tools spanning effects, cleanup, mastering, editing, labels, tracks, transcription and one-call pipelines. Everything runs locally through Audacity — no cloud, no API keys for the audio processing.
- Restoration and dynamics: get_noise_profile then noise_reduction, click_removal, truncate_silence, normalize, compressor, limiter and loudness_normalize
- Nine one-call cleanup pipelines that chain the steps for a job — auto_cleanup_podcast, auto_audiobook_mastering, auto_cleanup_interview, auto_cleanup_vocal, auto_cleanup_live, auto_master_music, auto_lofi_effect, auto_cleanup_audio and auto_analyze_audio
- Thirty effect_ tools — amplify, fade_in/fade_out, reverb, echo, change_pitch, change_tempo, change_speed, equalization, high_pass_filter, low_pass_filter, notch_filter, vocal_reduction, paulstretch, auto_duck and more
- Editing that mirrors Audacity: edit_cut, edit_copy, edit_paste, edit_split, edit_trim, edit_silence, edit_duplicate and the split-cut/split-delete variants
- Nineteen label_ tools — add, add_at, list, find, edit, import/export, regular_intervals, and region operations like label_cut_regions, label_silence_regions, label_export_chapters and label_export_audio_segments
- Track management: track_add_mono / track_add_stereo, track_mix_and_render, track_mute / track_unmute_all, track_resample, track_stereo_to_mono and track_align_end_to_end
- Analysis and generation: analyze_find_clipping, analyze_plot_spectrum, analyze_beat_finder, analyze_label_sounds, and generate_tone, generate_noise, generate_chirp, generate_dtmf and generate_rhythm_track
- Transcription that runs offline via faster-whisper — transcribe_audio, transcribe_selection, transcribe_to_labels (labels placed at each timestamp), transcribe_to_file, and transcription_set_model
- Project and transport control — project_new/open/save/import_audio/export_audio, plus transport_play, transport_record and cursor and selection tools
- check_pipeline_status reports how a running pipeline is progressing
A local Audacity 3.x installation (4.x is not yet supported) with mod-script-pipe enabled under Preferences → Modules, and Python 3.10 or later. The bundled installer installs audacity-mcp from the cloned folder, enables mod-script-pipe, and configures the client, with a --dry-run to preview. Transcription runs entirely offline through faster-whisper, so audio never leaves the machine.
One command — pip install audacity-mcp-server