Master a Suno Track from Claude or ChatGPT in One Message (MCP)

A raw Suno export usually sits at −18…−22 LUFS: on Spotify it sounds quieter than the tracks around it, and the file still carries the generator's digital fingerprint. The old fix was manual: download, open a site, pick a preset, wait, download again. Now your AI agent can do all of it — in the same chat where you wrote the lyrics.

Magic Master runs as a hosted MCP server. MCP is the standard Claude, ChatGPT, Cursor and other clients use to plug in outside tools. One address, nothing to install: https://magicmaster.pro/mcp.

Step 1. Connect (one minute)

Claude (claude.ai or Claude Desktop). Settings → Connectors → Add custom connector → URL https://magicmaster.pro/mcp/oauth. A Magic Master sign-in page opens — that is how the agent gets access to your tokens and trial masters.

ChatGPT. Developer mode → Connectors → create a connector with the same URL https://magicmaster.pro/mcp/oauth.

Cursor and VS Code. The page magicmaster.pro/mcp has Add to Cursor and Install in VS Code buttons — one click.

Claude Code. One line in the terminal:

claude mcp add --transport http magicmaster https://magicmaster.pro/mcp

Every recipe, including clients that only speak stdio, lives in github.com/denisok-ai/magicmaster-mcp and in the npm package magicmaster-mcp.

Step 2. Put the track behind a link

The agent fetches audio by direct link. Easiest: Dropbox (dl=0 → dl=1), a Google Drive file shared with "anyone with the link", your site or any cloud. Formats: wav, mp3, flac, m4a.

Step 3. Write one message

For example:

Here is my Suno track: https://…/night-drive.wav. Check the loudness, remove the Suno fingerprint, master it for Spotify and show me what you got.

What the agent does next — on its own, no clicks from you:

1. analyze_track — measures LUFS, true peak, duration, channel correlation and suggests a genre. Free.
2. clean_ai_trace — removes the Suno/Udio digital fingerprint without touching loudness. Free.
3. master_track — picks a preset for the genre and a loudness target (for Spotify that is −14 LUFS, true peak no higher than −1 dBTP) and starts the master. One token.
4. get_job — 20–60 seconds later collects the result and shows the mastering passport right in the chat: loudness before and after, true peak, dynamic range and how far it landed from the target.

If the agent hesitates between two styles it can compare short previews for free and only then spend the token.

What you get

A link to the finished file (wav, wav24, flac, mp3, opus or aac — your call) and an explanation of the numbers: why −14 rather than "as loud as possible", what true peak is, and why a −9 LUFS track gets turned down by the platform anyway. The rules the agent follows are public — they live in magicmaster.pro/agents.md.

Targets by destination

Destination	Target, LUFS	True peak
Spotify, YouTube, Deezer	−14	≤ −1 dBTP
Apple Music, podcasts	−16	≤ −1 dBTP
Audiobook	−19	≤ −1 dBTP
Club, DJ set	−9	≤ −1 dBTP

The agent picks the target from the destination, not from taste — that is written into its rules.

Pricing

• Analysis, fingerprint removal, status and download — free.
• Mastering — one token per full PRO master (any of 24 presets, every module, any format). Tokens never expire.
• Without an account — three masters a month. A registered agent gets three trial masters with no card.
• A human pays: the agent prepares the checkout, you confirm. It cannot pay by itself — on purpose.

When something goes wrong

The agent understands the error codes and will not loop: on an exhausted quota it names the reset time and offers a top-up, on a bad file it asks for another one, and if a master fails on our side the token is refunded automatically.

Connect: magicmaster.pro/mcp.