Audio mastering for AI agents: LUFS/True Peak targets, Suno/Udio AI-fingerprint removal.
Magic Master MCP — Audio Mastering for AI Agents
Remote MCP server that provides audio mastering services to MCP clients. It targets LUFS loudness and True Peak limiting, removes Suno/Udio AI fingerprints, and returns a “mastering passport” within the chat session. The server is accessible over streamable HTTP and uses defined protocol dates.
🛠️ Key Features
Audio mastering with LUFS target control
True Peak limiting
Suno/Udio AI-fingerprint removal
“Mastering passport” response in chat
Server endpoint: https://magicmaster.pro/mcp
🚀 Use Cases
Master audio intended for AI agents
Post-processing tracks to meet LUFS and True Peak targets
De-fingerprinting audio associated with Suno/Udio
⚡ Developer Benefits
Works with MCP clients (e.g., Claude, ChatGPT, Cursor, VS Code, Claude Code)
Streamable HTTP access
OAuth endpoint for paid tools: https://magicmaster.pro/mcp/oauth
Registry entry: pro.magicmaster/mastering
⚠️ Limitations
Does not specify supported input/output formats, tool names, or configuration details beyond LUFS/True Peak and fingerprint removal
All mastering presets with their target loudness in LUFS. Call this to get the exact preset keys that master_track and batch_master accept. Use preview_styles instead when you want to HEAR 2-6 of them on the actual track before choosing, and get_service_info when you need the wider service manifest rather than the preset list. Free.
What this account may do right now: token balance, free quota left and when it resets, upload cap, whether PRO modules are available. Call before mastering.
Measure integrated LUFS, sample peak, duration, sample rate and stereo correlation without processing. Free, a couple of seconds. Run this before choosing a preset. Note: true peak and loudness range are NOT part of this fast pass — they are measured on the result and come back from get_job.
Parameters1
audio_url
string
required
Public https URL of the audio file (wav/mp3/flac/m4a).
Master the middle 24 s of the track in 2-6 presets at once and get each one's loudness plus a listen link. FREE (6 per day). Use this instead of guessing a preset, and instead of burning a token to 'try' one.
Parameters2
audio_url
string
required
Public https URL of the audio file (wav/mp3/flac/m4a).
styles
array
optional
Preset keys from list_presets. Default: edm, pop, rnb, standard.
Master one track. COSTS 1 TOKEN when the account has a balance; otherwise it runs on the free quota without PRO modules. Returns a job id immediately — poll get_job (typical 20-60 s, up to 10 min with restoration). The source is fetched by this server, so the ceiling here is 100 MB per file even when the account allows more. Pass dry_run=true first if you want the price and the effective settings before anything is spent.
Parameters12
audio_url
string
required
Public https URL of the audio file (wav/mp3/flac/m4a).
style
string
optional
Preset key from list_presets, e.g. edm, hiphop, podcast, unlimiter. Default standard.
target_lufs
number
optional
Override the preset target, e.g. -14 for Spotify.
out_format
string
optional
bitrate
integer
optional
MP3: 128/192/256/320; OPUS: 128/192.
remove_watermark
boolean
optional
Also remove the Suno/Udio AI fingerprint. Free.
webhook_url
string
optional
Public https URL to POST the result to instead of polling.
restore_dynamics
boolean
optional
Open up a brickwalled source before mastering (pairs with style unlimiter).
intensity
number
optional
Multiplier on the preset's multiband compression: 1.0 as designed, 0.5 gentler, 1.5 denser. Values outside 0.5-1.5 are clamped.
denoise_strength
number
optional
Noise reduction, 0-1. 0 disables.
dither_type
string
optional
Dither for 16-bit WAV output: tpdf (default), ns_e or ns_itu (noise-shaped).
dry_run
boolean
optional
Do not master: report what this call would cost and what settings would apply. Nothing is charged, the file is not even downloaded.
Raw schema
{
"type": "object",
"properties": {
"audio_url": {
"type": "string",
"description": "Public https URL of the audio file (wav/mp3/flac/m4a)."
},
"style": {
"type": "string",
"description": "Preset key from list_presets, e.g. edm, hiphop, podcast, unlimiter. Default standard."
},
"target_lufs": {
"type": "number",
"description": "Override the preset target, e.g. -14 for Spotify."
},
"out_format": {
"type": "string",
"enum": [
"wav",
"wav24",
"mp3",
"flac",
"opus",
"aac"
]
},
"bitrate": {
"type": "integer",
"description": "MP3: 128/192/256/320; OPUS: 128/192."
},
"remove_watermark": {
"type": "boolean",
"description": "Also remove the Suno/Udio AI fingerprint. Free."
},
"webhook_url": {
"type": "string",
"description": "Public https URL to POST the result to instead of polling."
},
"restore_dynamics": {
"type": "boolean",
"description": "Open up a brickwalled source before mastering (pairs with style unlimiter)."
},
"intensity": {
"type": "number",
"minimum": 0.5,
"maximum": 1.5,
"description": "Multiplier on the preset's multiband compression: 1.0 as designed, 0.5 gentler, 1.5 denser. Values outside 0.5-1.5 are clamped."
},
"denoise_strength": {
"type": "number",
"minimum": 0,
"maximum": 1,
"description": "Noise reduction, 0-1. 0 disables."
},
"dither_type": {
"type": "string",
"enum": [
"tpdf",
"ns_e",
"ns_itu"
],
"description": "Dither for 16-bit WAV output: tpdf (default), ns_e or ns_itu (noise-shaped)."
},
"dry_run": {
"type": "boolean",
"description": "Do not master: report what this call would cost and what settings would apply. Nothing is charged, the file is not even downloaded."
}
},
"required": [
"audio_url"
],
"additionalProperties": false
}
batch_master
Master up to 10 tracks in one call with the same settings. Needs N TOKENS for N tracks, or N free runs — not one. album_mode (default true) levels the tracks RELATIVE to each other: the loudest hits the target exactly, the others keep their offsets, so a quiet interlude is not inflated to single level. Limits: 10 tracks, 100 MB per file and 300 MB per call (this server's own caps, for memory) — the account may allow bigger uploads on the site, but not through this channel. A big album takes minutes — send fewer, longer tracks per call rather than all at once. Returns one job id per track — poll each with get_job. Pass dry_run=true to see the price and the settings before anything is spent.
Parameters8
audio_urls
array
required
Public https URLs, in album order.
style
string
optional
Preset key from list_presets.
target_lufs
number
optional
Target for the LOUDEST track in album mode.
out_format
string
optional
bitrate
integer
optional
MP3: 128/192/256/320; OPUS: 128/192.
remove_watermark
boolean
optional
Also remove the Suno/Udio AI fingerprint. Free.
album_mode
boolean
optional
Keep relative loudness between tracks. Default true. Set false for unrelated singles.
dry_run
boolean
optional
Report the price and settings without mastering and without downloading anything.
Raw schema
{
"type": "object",
"properties": {
"audio_urls": {
"type": "array",
"minItems": 1,
"maxItems": 10,
"items": {
"type": "string"
},
"description": "Public https URLs, in album order."
},
"style": {
"type": "string",
"description": "Preset key from list_presets."
},
"target_lufs": {
"type": "number",
"description": "Target for the LOUDEST track in album mode."
},
"out_format": {
"type": "string",
"enum": [
"wav",
"wav24",
"mp3",
"flac",
"opus",
"aac"
]
},
"bitrate": {
"type": "integer",
"description": "MP3: 128/192/256/320; OPUS: 128/192."
},
"remove_watermark": {
"type": "boolean",
"description": "Also remove the Suno/Udio AI fingerprint. Free."
},
"album_mode": {
"type": "boolean",
"description": "Keep relative loudness between tracks. Default true. Set false for unrelated singles."
},
"dry_run": {
"type": "boolean",
"description": "Report the price and settings without mastering and without downloading anything."
}
},
"required": [
"audio_urls"
],
"additionalProperties": false
}
get_job
Status, progress and result metrics (after_lufs, true_peak_db, lra, platform_ready) for a mastering, cleanup or preview job. For previews it returns each style with its loudness and a listen link. When a master is done it also returns download_url — a signed link that works without credentials for about an hour, so an OAuth-connected agent can fetch the file it paid for. Free.
Parameters1
job_id
string
required
Job id returned by master_track, clean_ai_trace or preview_styles; batch_master returns one per track in job_ids. Not a URL or a filename: 1-64 chars of A-Z a-z 0-9 _ -
Raw schema
{
"type": "object",
"properties": {
"job_id": {
"type": "string",
"description": "Job id returned by master_track, clean_ai_trace or preview_styles; batch_master returns one per track in job_ids. Not a URL or a filename: 1-64 chars of A-Z a-z 0-9 _ -",
"minLength": 1,
"maxLength": 64,
"pattern": "^[A-Za-z0-9_-]{1,64}$"
}
},
"required": [
"job_id"
],
"additionalProperties": false
}
create_topup_link
Prepare a token purchase for the human principal. The agent CANNOT complete the payment: card and Telegram Stars both need the human. Returns the checkout link to hand over, then poll check_limits.
Parameters1
pack_id
string
optional
Pack id from get_pricing, e.g. tokens_10.
Raw schema
{
"type": "object",
"properties": {
"pack_id": {
"type": "string",
"description": "Pack id from get_pricing, e.g. tokens_10."
}
},
"additionalProperties": false
}
Remote MCP server that lets Claude, ChatGPT, Cursor, VS Code, Claude Code, OpenAI Codex and any other MCP client master audio: hit a loudness (LUFS) target with True Peak limiting, remove the Suno / Udio AI fingerprint, compare presets on a 24-second preview, and master a whole album with consistent relative loudness. Results come back with measured numbers and a mastering passport in the chat.
Service version: 3.15.0 (as reported by get_service_info)
Nothing to install: the server is hosted. This repository holds the connection recipes and a small stdio bridge for clients that cannot speak HTTP.
Tools
The server exposes 11 tools. Descriptions below are condensed from the live tools/list; the live text is authoritative.
Tool
What it does
Cost
get_service_info
What Magic Master does, what is free, what costs a token, and where the rules live. Call once at the start of a session.
free
list_presets
All 24 mastering presets with their target loudness in LUFS. Gives the exact keys that master_track and batch_master accept.
free
get_pricing
Live token packs and payment methods (RUB). Prices are read live, not quoted from memory.
free
check_limits
Token balance, free quota left and when it resets, upload cap, whether PRO modules are available.
free
analyze_track
Integrated LUFS, sample peak, duration, sample rate and stereo correlation, without processing. True peak and loudness range come back from get_job after mastering.
free
clean_ai_trace
Strips the Suno / Udio AI generator fingerprint; loudness and tone stay as they are. Returns a job id.
free
preview_styles
Masters the middle 24 seconds in 2–6 presets at once and returns each one's loudness and a listen link.
free, 6 per day
master_track
Masters one track. Costs 1 token when the account has a balance; otherwise it runs on the free quota without PRO modules. Returns a job id. Pass dry_run=true to see the price and effective settings first.
1 token
batch_master
Masters up to 10 tracks in one call with the same settings. album_mode (default true) keeps relative loudness between tracks. Returns one job id per track.
1 token per track
get_job
Status, progress and result metrics (after_lufs, true_peak_db, lra, platform_ready) for a mastering, cleanup or preview job. Includes a signed download_url when a master is done.
free
create_topup_link
Prepares a token checkout for the human. The agent cannot complete the payment itself.
your client cannot send custom headers (Claude Desktop, ChatGPT connectors)
Without any key, the free tools work on /mcp.
On /mcp, Authorization: Bearer means an OAuth access token. An account JWT sent there returns 401, and it breaks even the free calls. Use X-API-Key on /mcp, or connect through /mcp/oauth.
/mcp/oauth answers an unauthenticated request with 401 and a WWW-Authenticate challenge pointing at https://magicmaster.pro/.well-known/oauth-protected-resource, so a compliant client starts the sign-in on its own.
OAuth metadata: /.well-known/oauth-authorization-server; dynamic client registration: /oauth/register. Access tokens last one hour; refresh tokens rotate on every use. Tokens are issued for the resource https://magicmaster.pro/mcp only.
API keys (mm_…) are issued in the profile after a token purchase; up to 10 active keys.
Install
Claude Code
bash
claude mcp add --transport http magicmaster https://magicmaster.pro/mcp
Free tools work immediately. For the paid tools, add your key:
bash
claude mcp add --transport http magicmaster https://magicmaster.pro/mcp --header "X-API-Key: mm_YOUR_KEY"
Claude Desktop and claude.ai (custom connector, OAuth)
Settings → Connectors → Add custom connector → URL https://magicmaster.pro/mcp/oauth → sign in on the consent page. Use https://magicmaster.pro/mcp if you only need the free tools.
Claude Desktop through a stdio bridge (if you prefer the config file)
The bridge points at https://magicmaster.pro/mcp/oauth by default. Set MAGICMASTER_MCP_URL=https://magicmaster.pro/mcp to use the anonymous endpoint instead.
One-click buttons: Add to Cursor · VS Code: vscode:mcp/install?{"name":"magicmaster","type":"http","url":"https://magicmaster.pro/mcp"}
Docker
For clients without Node, and for the Glama build that runs security checks:
bash
docker build -t magicmaster-mcp .
docker run -i --rm magicmaster-mcp
The image holds only the bridge; the server itself stays hosted at https://magicmaster.pro/mcp. It is built from this repository (npm ci), so mcp-remote is baked in and the container needs only outbound HTTPS to magicmaster.pro at run time. The image defaults to the anonymous endpoint. To use OAuth (which also unlocks the paid tools), run it with -e MAGICMASTER_MCP_URL=https://magicmaster.pro/mcp/oauth.
Build spec for a Glama release (the platform configures the recipe on its side, it does not read this Dockerfile):
Free, no account:analyze_track, clean_ai_trace, preview_styles (6 per day), get_job, get_pricing, list_presets, get_service_info, check_limits, create_topup_link.
Guest mastering (no account): 3 basic masters per calendar month per IP. Basic means no PRO modules.
Free account: 1 master per UTC day (3 during the first 24 hours after sign-up). Also basic.
Trial for agents: registering an agent once (POST /api/agents/register, with the account's JWT as Authorization: Bearer) grants 3 full PRO masters and an API key. The MCP connector does not grant the trial; only that call does. See agents.md.
Tokens: 1 token = 1 full PRO master (any preset, all PRO modules, any export format). Tokens never expire. There are no subscriptions. Some packs include bonus tokens. Live packs and prices: get_pricing (RUB) or GET https://magicmaster.pro/api/tokens/packages.
Payment needs the human. An agent cannot complete a card payment. Call create_topup_link, hand the checkout link to the person, then poll check_limits until the balance is credited. A card issued outside Russia may be declined; see https://magicmaster.pro/wallet.
master_track and batch_master accept dry_run=true, which reports the price and the effective settings without spending anything.
Formats, limits and retention
Input (from agent.json): aac, aif, aiff, amr, flac, m4a, mp3, ogg, wav, webm. Detection is by content.
Upload caps: on the site and API, 160 MB without payment; after a token purchase up to 1000 MB (WAV). Through this MCP server, master_track and batch_master accept up to 100 MB per file and batch_master up to 300 MB per call (10 tracks). check_limits returns the effective per-channel ceiling (max_upload_mb_via_mcp).
Audio source: a public https URL (audio_url). Share links to Suno/Udio song pages, Google Drive and Dropbox are resolved to the file.
Retention of finished masters: guest results are kept only briefly (in memory); free accounts 1 day; after a token purchase 30 days. get_job returns a signed download_url for about an hour, which works without credentials.
Typical time: 20–60 s per master; up to 10 minutes for a long track with restoration. Poll get_job, or pass webhook_url to master_track to get a POST instead.
Mastering options
Presets (24, from list_presets): e.g. standard (−14 LUFS), edm (−9), pop (−12), hiphop (−13), podcast (−16), unlimiter (−12), voice (−16), audiobook (−19). Use preview_styles to compare two to six of them on 24 seconds for free.
Target loudness:target_lufs overrides the preset, e.g. -14 for Spotify. True Peak limiting follows ITU-R BS.1770-5.
Output:out_format, bitrate, dither_type (tpdf, ns_e, ns_itu, for 16-bit WAV).
Density:intensity from 0.5 (gentler) to 1.5 (denser); 1.0 is the preset as designed.
Restoration:restore_dynamics opens up a brickwalled source before mastering (pairs with the unlimiter preset). denoise_strength (0–1) sets noise reduction; 0 disables it. The web restoration page also lists de-hum and de-click steps, which the MCP tools do not expose by name.
Suno / Udio cleanup:remove_watermark=true on master_track and batch_master also removes the AI fingerprint, free of charge. The standalone tool is clean_ai_trace.
Album mode:batch_master with album_mode (default true): the loudest track hits the target and the others keep their offsets, so a quiet interlude is not inflated to single level. Use album_mode=false for unrelated singles.
What is not in MCP
These features are on the website (or in the API) but are not MCP tools:
Stem mastering (2–8 stems combined and mastered as one track): API POST /api/v2/master/stems, 1 token.
Tool list documented as the live 11 tools: adds preview_styles (free comparison, 6 per day) and batch_master (album mode, up to 10 tracks per call); both were missing from the 0.3.0 README.
README: Codex, Claude Code with API key header, and OAuth details (/mcp/oauth, PKCE, scope mastering) added; limits and formats taken from the live manifest.
Bridge (bin/magicmaster-mcp.js) uses the bundled mcp-remote dependency when it is installed; it falls back to npx only when it is not.
Package metadata: description and keywords extended.
По-русски
Удалённый MCP-сервер Magic Master для AI-агентов и клиентов вроде Claude, ChatGPT, Cursor, VS Code, Claude Code и Codex. Он выводит трек на целевую громкость (LUFS) с ограничением True Peak, снимает цифровой след Suno/Udio, сравнивает пресеты на 24-секундном отрывке и мастерит целый альбом с сохранением относительной громкости. Ставить ничего не нужно — адрес https://magicmaster.pro/mcp.
Бесплатно: анализ, снятие следа ИИ, сравнение пресетов (6 раз в сутки), статус задач, цены и лимиты.
Платно: 1 токен = 1 полный PRO-мастеринг. Токены не сгорают, подписок нет. Актуальные цены — инструмент get_pricing (в рублях). Оплату проводит человек по ссылке из create_topup_link.
Подключение: Claude Code — claude mcp add --transport http magicmaster https://magicmaster.pro/mcp; Claude Desktop и claude.ai — пользовательский коннектор на https://magicmaster.pro/mcp/oauth.