aidaw
This server is an MCP backend for AI-driven music production: project editing, rendering, plugin/catalog discovery, audio analysis, and delivery.
Manage DAW projects: create/list/inspect projects, apply atomic musical edits, restore revisions, and import/export portable project data.
Render audio: start offline WAV rendering of projects or individual tracks/stems, batch render, poll/cancel render jobs, and export MIDI/project files.
Analyze audio: measure peak/RMS/silence/clipping, loudness/true peak, inspect MIDI timing/notes, and inspect delivery manifests.
Discover and manage plugins: scan/inventory/verify VST3/AU plugins, inspect parameters/programs, save/import presets, probe instrument/effect sounds, and audition presets in project context.
Search and assess sounds/effects: search sound/effect records, store caller/human assessments and parameter annotations, and run knowledge searches across probes/assessments/feedback.
Index content libraries: discover/register sample/preset roots, index/search library patches, and bind content to saved presets and probe audio.
Publish/deliver: inspect/publish delivery manifests, bundle/export/import project packages, relink assets, and monitor the server processing queue.
Allows importing and inspecting Standard MIDI Files, converting notation into MIDI parts, and exporting performance/notation MIDI for music production.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@aidaw Create a new project with a synth track and render a 30-second demo."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
AIDAW

AIエージェントへ: 作業は AGENTS.md から該当する手順へ進んでください。このREADMEは人向けの紹介です。README自体の編集・説明時は対象本文を読んでください。
作曲・マスタリング・ステム分離を、AIDAWひとつで。
思い描いた曲を、言葉から音へ。仕上げも、パート分けも、AIDAWで。
AIDAWは、無料で使えるオープンソースの音楽制作ソフトです。 AIDAW本体の利用料はかかりません。詳しくは 利用規約・ライセンス案内 をご覧ください。
AIDAWは、AIに話しかけて音楽を作る Vibe Coding DAW。曲を作る標準音源、MTSで採用していたGoldenTips相当のマスタリング、音声を4パートに分けるステム分離を、最初から備えています。
別のDAWや有料プラグイン、分離専用アプリをそろえずに、制作から試聴・書き出しまで。編集はAIに頼み、再生や聴き比べは操作画面 AIDAW DECK で直接操作。専用コントローラーも必要ありません。
「曲を作りたい」「完成した曲を仕上げたい」「ボーカルと伴奏を分けたい」。その目的に合ったプロジェクトを選んで、すぐに制作を始められます。
Related MCP server: reaper-mcp-server
はじめ方
ローカルのファイル操作とコマンド実行ができる Codex・Claude Code・Antigravityのいずれか に、次の文章を送ってください。
https://github.com/momuandteasteam/aidaw.git をクローンして、
AGENTS.mdに従ってAIDAWをセットアップしてください。
今使っているAIクライアントからの接続を確認し、AIDAW DECKを起動してください。クローン済みなら、そのフォルダをAIで開いて「AIDAWをセットアップして」で始められます。AIがビルド、標準音源・エフェクト・分離モデルの導入、接続設定を進めます。認証・信頼確認が求められたら案内に従い、必要に応じてAIクライアントを再起動してください。
準備ができたら、以下の3つからやりたいことを依頼してください。コード枠内はAIに送る依頼文です。音声を使う場合は、ファイルを添付するか保存先を伝えます。
曲を作る
「落ち着いたピアノ曲を」「ベースをもっと前に」。作りたい音楽をAIに伝えて、標準音源 AIDAW GM で曲づくり。MIDIとオーディオのトラックを組み合わせ、演奏・音色・ミックスを調整できます。
「夜の散歩」という曲制作プロジェクトを作成してください。
90 BPMの落ち着いた曲を、ピアノ・ベース・ドラムでまず8小節。
標準音源のAIDAW GMを使って、AIDAW DECKで聴けるようにしてください。聴いた後は「ドラムを少し控えめに」「ピアノにリバーブを足して」と伝えれば、さらに調整できます。手持ちの対応音源やエフェクトも名前で指定できます。
マスタリングする
完成した曲を読み込み、音色・広がり・低域・音圧を整えます。GoldenTips方式を基準にしたEnhancerと、8バンドEQ・Imager・BassMono・Limiterなどの標準エフェクトで、1曲からアルバムまで。変更前後をA/Bで聴き比べ、納得した音を選べます。
マスタリングプロジェクト「ミニアルバム」を作成して、添付した3曲を読み込んでください。
曲ごとの音量感と音色を整え、変更前後をA/Bで聴き比べられるようにしてください。GoldenTipsと同一の出力を保証するものではありません。処理方式の詳細は 標準エフェクトDSP を参照してください。
ステム分離する
音声を ボーカル・ドラム・ベース・その他 の4パートへ。別の分離アプリやサービスを用意せず、パートごとに聴いたり、ボーカルを外して伴奏にしたりできます。全ステムのZIP保存にも対応しています。
ステム分離プロジェクトを作成して、添付した音声を分離してください。
全ステムをZIPで保存し、ボーカルをミュートした伴奏もWAVで書き出してください。標準は Spleeterの4ステムモデルです。分離は推定処理のため、他パートの音が残る場合があります。セットアップで導入するDemucsを含め、方式とモデルの利用条件は ステム分離 を参照してください。
AIDAW DECKで聴く・操作する
再生・停止・シーク・試聴・読み込み・書き出しを、手元で。 AIへの編集依頼と組み合わせて使うコントロール用GUIです。開いたプロジェクトはAIと共有され、編集前に対象を確認します。
AIDAW DECKはセットアップで導入されます。macOSはアプリケーション、Windowsはスタートメニューから起動できます。本体の更新後は、AIに「AIDAW DECKを更新して」と依頼してください。操作の詳細は 再生・試聴・書き出し をご覧ください。
書き出す・保存する
音声は WAV・MP3・FLAC で書き出せます。曲制作のトラック別ステム、マスタリングの曲単位・アルバム全体にも対応しています。
この曲をWAVとMP3で書き出してください。
編集を続けられるように、プロジェクトも.aidaw.zipでエクスポートしてください。普段の作品は Projects/ に保存されます。.aidaw.zip は編集データと履歴を持ち出すための形式です。以前の状態に戻したいときは、AIに「編集履歴を確認して、直前の変更前に戻して」と依頼できます。
使える音源・エフェクト・ファイル
標準搭載は AIDAW GM、Gain、EQ(8バンド)、Limiter、Reverb、Enhancer、Imager、BassMono。手持ちの VST3、macOSの Audio Units/AU も利用できます。インストール・認証済みの製品を名前で指定してください。すべての製品・プリセット形式への対応は保証していません。VST2は対象外です。
用途 | 対応形式 |
音声の読み込み | WAV・AIFF・FLAC・Ogg Vorbis。MP3などはOS・デコーダーに依存します。 |
音声の書き出し | WAV(24 bit)・MP3(320 kbps)・FLAC。48 kHz・ステレオ。 |
演奏データ | MIDIの読み込み・書き出し。 |
編集用プロジェクト |
|
音声処理は 48 kHz です。異なるサンプルレートや読み込めない形式の素材は、AIに「原音を残して48 kHzのWAVへ変換してから読み込んで」と依頼してください。
対応環境・詳しい手順
対応OS:Windows 11・macOS・Linux(未検証)。現在は開発中です。機種専用コントローラー連携などの対応範囲は 要件と対応状況 をご覧ください。
正式なロゴ・アイコンは ブランド素材 をご覧ください。
開発に参加する
AIと直した不具合や追加機能は、「この変更を作者に共有して、プルリクを作成して」と依頼してAIDAWへ還元できます。AIが変更とテストを確認し、GitHubにレビュー用の提案を送る準備を進めます。初回はGitHubへのログインが必要です。作品・音声・個人設定は共有対象に含めません。
Available Tools
63 toolsasset_importC
Copy an immutable source/reference/artwork into the project, hash and analyze it. Reference audio is forbidden as render input.
Inputs: path (required), role (required), project_id (required).
asset_listC
List project assets and provenance.
Inputs: project_id (required).
asset_relinkA
Restore a missing/corrupt collected asset from a file with the exact original hash. Keeps source/reference role unchanged.
Inputs: path (required), asset_id (required), project_id (required).
audio_analyzeA
Measure a local mono/stereo WAV: sample peak, RMS, silence and clipping. This does not measure LUFS or true peak.
Inputs: path (required).
audio_measureA
Measure source integrated LUFS, true peak and loudness range with FFmpeg. Does not normalize or edit the source, or prescribe -14 LUFS.
Inputs: path (required).
batch_delivery_publishA
Publish all completed batch songs together as WAV/optional tagged MP3/report/ZIP using one recoverable release manifest.
Inputs: job_id (required), project_id (required), mp3, tags.
batch_renderC
Master multiple source assets in ONE project job. No implicit cut; failed tasks can be resumed without repeating completed files.
Inputs: tasks (required), project_id (required), wait.
batch_render_resumeC
Resume a source-audio batch using its pinned snapshot and task settings.
Inputs: job_id (required), project_id (required), wait.
bundle_exportA
Collect current project, hashed assets, and frozen audio from a successful current-revision render. Excludes jobs, temp and existing ZIPs.
Inputs: job_id (required), project_id (required).
bundle_importB
Import a validated portable package under a new project ID; never substitute missing plugins. Includes frozen playback.
Inputs: path (required), project_id (required).
catalog_discoverC
List plugin candidate locations without loading them. Scan individual candidates next.
Inputs: format (required), limit, offset, search_path.
catalog_feedbackC
Store user preference for one exact probe/assessment, not a global plugin prohibition.
Inputs: comment (required), reviewer (required), record_id (required), preference (required).
catalog_indexA
Incrementally record local plugin/preset metadata without declaring render compatibility.
Inputs: no named parameters.
catalog_inventoryB
Discover every candidate in the requested formats, scan each in isolation, index exposed parameters/programs, and checkpoint one resumable job. Failed plugins do not stop the remaining inventory.
Inputs: formats, max_seconds.
catalog_inventory_resumeB
Resume pending candidates in an existing plugin inventory job.
Inputs: job_id (required), max_seconds.
catalog_inventory_statusC
Read plugin inventory coverage and per-candidate failures without running plugin code.
Inputs: job_id (required).
catalog_portable_exportA
Export a sanitized metadata catalog without installation paths, plugin state, binaries or licenses.
Inputs: job_id.
catalog_reference_searchA
Search the source-controlled observed metadata catalog. Results describe known plugin parameters/programs but do not prove installation on this server.
Inputs: limit, query.
catalog_scanC
Scan ONE candidate in an isolated worker and register its plugins. Scan success is not render compatibility certification.
Inputs: format (required), location (required).
catalog_searchB
Find installed plugins and saved presets. Basic GM/builtin candidates are excluded unless include_basic is explicit. Tags on saved presets are user-provided.
Inputs: limit, query, include_basic, instrument_only.
content_bind_presetA
Link a caller-identified library patch to an exact saved preset and its actual probe audio. Does not claim independent proof of patch identity.
Inputs: probe_id (required), preset_id (required), content_id (required), description (required).
content_discover_rootsA
Read NI registered content locations on Mac/Windows and known user/shared directories. Does not load instruments or infer licensing.
Inputs: no named parameters.
content_indexB
Index instrument patches INSIDE registered libraries (.nki/.nksn/.nksf/.nbkt/.nmsv/.ens etc). Bounded scan; reports gaps, not false completeness.
Inputs: root_ids, max_files, max_seconds.
content_register_rootB
Add an explicit third-party sample library/preset root without moving or changing its files.
Inputs: name (required), path (required), family.
content_searchC
Find library/patch-level candidates and host plugin links. Discovered files are NOT certified loadable or auditioned.
Inputs: limit, query, family, offset.
delivery_inspectA
Read the already published delivery manifest and server file paths without rendering or conversion.
Inputs: project_id (required).
delivery_publishA
Validate and publish a current render with stems, optional tagged MP3, MIDI, report and delivery ZIP. Journals updates for rollback. Does not certify listening quality.
Inputs: tags (required), job_id (required), project_id (required), mp3, +1 more.
delivery_recoverB
Recover an interrupted output publication from its journal.
Inputs: project_id (required).
effect_indexC
Read and store plugin parameters and programs. Parameter semantics remain unmapped until supported by evidence.
Inputs: plugin (required).
effect_probeC
Render an effect on a short input and return before/after audio for the caller to assess.
Inputs: path (required), plugin (required).
effect_searchC
Find effect records with parameter/effect evidence and current registration.
Inputs: limit, query.
export_midiA
Write a Standard MIDI File type 1 with notes, tempo and meter. Plugin sounds/effects remain in the project, not MIDI.
Inputs: project_id (required).
export_notation_midiB
Export notation pitches without keyswitch notes, separately from plugin performance MIDI.
Inputs: project_id (required).
export_projectA
Export a portable project JSON with concrete notes and saved plugin states. Does not bundle plugin binaries or sample libraries.
Inputs: project_id (required).
instrument_probeC
Probe a bounded preset batch in one job; retry/resume preserves completed tasks.
Inputs: plugins (required), bpm, pitches, velocity, +1 more.
instrument_probe_resumeC
Resume failed/pending tasks in an existing probe batch.
Inputs: job_id (required).
job_cancelB
Cancel a render owned by this MCP service. Partial WAV files are discarded.
Inputs: job_id (required).
job_statusB
Read render progress, errors, finished output and audio measurements.
Inputs: job_id (required).
knowledge_searchC
Search stored parameters, probes, and assessments. Search results retain evidence type; unreviewed probes are not AI listening results.
Inputs: kind, limit, query.
massive_x_preset_importB
Import a local NKS preset for Massive X VST3 1.7.1 (R0), validating its product ID and capturing plugin state. Requires an installed and initialized plugin; does not bundle its content.
Inputs: name (required), path (required), plugin (required), tags.
midi_importC
Import selected MIDI parts with explicitly assigned instruments in one revision. Unsupported events require explicit notes_only acknowledgment.
Inputs: path (required), tracks (required), project_id (required), request_id (required), +3 more.
midi_inspectA
Parse SMF 0/1 notes and timing. Report unsupported controllers/program changes; variable tempo is rejected.
Inputs: path (required).
mixer_inspectC
Read track strips, pre/post sends, FX return strips and master chain. No GUI is required.
Inputs: project_id (required).
modo_bass_preset_importA
Import a local .mb2 musical preset using the version-checked MODO BASS 2 VST3 2.0.5 adapter. Returns a reusable preset ID after state readback verification.
Inputs: name (required), path (required), plugin (required), tags.
parameter_annotateA
Attach evidenced meaning to a known parameter snapshot; never infer a linear normalized-to-unit mapping.
Inputs: meaning (required), evidence (required), snapshot_id (required), parameter_id (required).
plugin_inspectC
Load a registered plugin and inspect its normalized parameter IDs and programs. No GUI is opened.
Inputs: plugin (required), limit, offset.
plugin_preset_saveC
Apply a program/normalized parameters and save a named state preset. Returns preset ID usable in a track instrument/effect.
Inputs: name (required), plugin (required), tags.
plugin_verifyC
Isolated load and state roundtrip test, with two actual renders for instruments. Records version/platform/binary and distinct untested capabilities.
Inputs: plugin (required).
project_applyA
Atomically apply concrete musical edits at base_revision. Use a unique request_id; identical retries are safe. Captures plugin state before committing.
Inputs: operations (required), project_id (required), request_id (required), base_revision (required).
project_cleanupA
List or remove only completed job work files. Never removes artifacts, source assets, frozen audio or failed jobs.
Inputs: project_id (required), dry_run.
project_createB
Create an editable project. Timing uses integer ticks, 960 ticks per quarter note. No implicit music generation.
Inputs: name (required), project_id (required), length_ticks (required), bpm, +3 more.
project_importC
Import exported project data under a new project ID. Plugins must be rescanned on the destination machine before rendering.
Inputs: project (required), project_id (required).
project_inspectA
Read current revision, sections, harmony and tracks. Notes are omitted by default; request them for precise edits.
Inputs: project_id (required), include_notes.
project_listA
List local projects.
Inputs: no named parameters.
project_restoreC
Restore a previous saved revision as a NEW revision, retaining history.
Inputs: revision (required), project_id (required), request_id (required), base_revision (required).
project_validateB
Check assets and plugin dependencies; return frozen playback when available. Does not claim live plugin compatibility.
Inputs: project_id (required).
queue_statusB
Show the single server processing lane and FIFO waiting operations. Connections and job reads stay responsive.
Inputs: no named parameters.
render_startB
Start offline WAV rendering of a fixed revision. Poll job_status; a succeeded job provides an absolute output path. track_id exports a stem with track inserts and without master effects.
Inputs: project_id (required), track_id, request_id, tail_seconds, +1 more.
sound_assessB
Append caller or human assessment tied to exact probe audio. Use metadata_inference if the audio was not listened to.
Inputs: tags (required), uses (required), method (required), probe_id (required), +4 more.
sound_audition_in_contextB
Audition a preset on an existing MIDI part in context without changing the project.
Inputs: plugin (required), track_id (required), project_id (required).
sound_probeC
Render a short preset audition and return a local audio path and hash for the calling AI to review. No model/provider configuration.
Inputs: plugin (required), bpm, pitches, velocity.
sound_searchB
Find sound records with caller assessments and currently registered instrument availability.
Inputs: limit, query.
system_capabilitiesA
Report the native engine platform and supported features. AU is macOS only; VST3 is shared with Windows.
Inputs: no named parameters.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 63 tools
Several tools occupy overlapping boundaries: project_import vs bundle_import, sound_probe vs instrument_probe vs effect_probe, catalog_scan vs catalog_inventory vs catalog_discover, and audio_measure vs audio_analyze. The descriptions are detailed, but the sheer number of closely related probe/search/index/import/publish tools would make selection genuinely difficult for an agent.
The majority of tools follow a consistent snake_case object_action pattern (project_import, sound_search, asset_list, render_start). The main deviations are the action-first export_* tools and noun-style status tools like queue_status, job_status, and catalog_inventory_status, but the overall convention remains predictable.
At 63 tools, this is far beyond the 25+ threshold and into the extreme range for a single MCP server. Even if each subdomain is individually useful, the combined surface is too large for coherent navigation and would be better split into separate project, catalog, and delivery servers.
The tool set covers a broad and coherent DAW workflow: project lifecycle, asset management, MIDI handling, plugin/catalog/content discovery, rendering, batch jobs, delivery, and audio measurement. Minor gaps exist, such as no explicit project deletion, asset deletion, or mixer modification tools, but the core workflows are well supported.
Maintenance
Related MCP Connectors
Remote MCP server for audio stem separation, denoise, reverb, and MIDI export.
AI image, video, voice and music generation over MCP, routed to Veo 3.1, Seedance 2.5 and more.
Read and edit DAW projects as an agent, save native Logic projects, convert projects between DAWs
MCP server for Producer/Riffusion AI music generation
Related MCP Servers
- AlicenseBqualityBmaintenanceA Model Context Protocol server that enables AI agents to create fully mixed and mastered tracks in REAPER DAW, supporting project management, MIDI composition, audio recording, and mixing automation.61185MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that lets language models interact with the Reaper DAW1MIT
- AlicenseNot gradedqualityBmaintenanceAI-powered music production toolkit enabling AI agents to generate, remix, separate stems, master tracks, and control DAWs via MCP.1MIT
- FlicenseAqualityDmaintenanceMCP server for controlling REAPER digital audio workstation from natural language. It enables AI agents to create tracks, import audio, adjust mix parameters, and manage sessions through typed operations and a REAPER bridge.22-