Audio MCP Server Documentation

Overview

The server speaks the Model Context Protocol over Streamable HTTP. Point any MCP client at the endpoint below and authenticate with a VocalRemover API token. Files are passed as public http(s) URLs (up to 100 MB) and results come back as downloadable URLs.

Endpoint:  https://vocalremover.com/mcp/audio
Transport: Streamable HTTP (POST)
Auth:      Authorization: Bearer YOUR_API_TOKEN

Quickstart

1. Get a free API token on the API page.
2. Add the server to your client using a snippet below.
3. Ask: "Remove the vocals from ."

Authentication

1. Create a free account and open the API page.
2. Generate an API token.
3. Add it as a Bearer token in your MCP client config (see below).

The free audio tools and the paid AI tools both use this token; the difference is only whether conversion-minute credits are charged.

Client setup

Add the server to your AI client. Claude Desktop uses the mcp-remote bridge; Cursor, Windsurf, Cline, Zed and Claude Code support remote MCP natively with a Bearer token; ChatGPT and the Claude.ai web app add it as a custom connector and sign in via OAuth (see below).

{
  "mcpServers": {
    "vocalremover": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote", "https://vocalremover.com/mcp/audio",
        "--header", "Authorization:${AUTH}"
      ],
      "env": { "AUTH": "Bearer YOUR_API_TOKEN" }
    }
  }
}

{
  "mcpServers": {
    "vocalremover": {
      "url": "https://vocalremover.com/mcp/audio",
      "headers": {
        "Authorization": "Bearer YOUR_API_TOKEN"
      }
    }
  }
}

claude mcp add --transport http \
  vocalremover https://vocalremover.com/mcp/audio \
  -H "Authorization: Bearer YOUR_API_TOKEN"

Anthropic Messages API (build your own agent)

Attach the server to a Claude API call with the MCP connector beta header:

client.beta.messages.create(
    model="claude-opus-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Remove vocals from https://example.com/song.mp3"}],
    mcp_servers=[{
        "type": "url",
        "url": "https://vocalremover.com/mcp/audio",
        "name": "vocalremover",
        "authorization_token": "YOUR_API_TOKEN",
    }],
    betas=["mcp-client-2025-11-20"],
)

ChatGPT (custom connector)

In ChatGPT settings, enable Developer mode, open Apps & Connectors → Create, and add a custom connector pointing at the server URL below. ChatGPT runs the OAuth flow — you sign in to VocalRemover and approve access; no token to paste. (Requires a plan with custom connectors, e.g. Plus/Pro/Business/Enterprise.)

https://vocalremover.com/mcp/audio

No custom-connector access on your plan? Create a Custom GPT instead and add an Action that imports our OpenAPI schema with API-key (Bearer) auth.

Claude.ai web app (custom connector)

In Claude.ai, go to Settings → Connectors → Add custom connector, paste the server URL, and connect. Claude runs the OAuth flow — sign in to VocalRemover and approve. (Available on Pro, Max, Team and Enterprise.)

https://vocalremover.com/mcp/audio

Tools

Uploading local files

Tools take a public URL. If the file is on the user's machine, call the free create_upload tool to get a one-time upload URL, POST the file to it (no token needed), then pass the returned file_url to any audio tool. The upload link and file expire after 1 hour. Never base64 a real audio file — it is far too large for a tool argument.

Assistant calls:  create_upload { "filename": "song.mp3" }
            ->  { "upload_url": "https://vocalremover.com/api/uploads/01h...",
                  "file_url":   "https://vocalremover.com/storage/mcp-uploads/01h....mp3",
                  "upload_page":"https://vocalremover.com/upload/01h..." }

# Upload the bytes (model's sandbox, a terminal, or the user's browser):
curl -F "file=@song.mp3" "https://vocalremover.com/api/uploads/01h..."
            ->  { "file_url": "https://vocalremover.com/storage/mcp-uploads/01h....mp3" }

Assistant calls:  separate_audio { "url": "", "task": "vocals" }

No way to make web requests (e.g. a sandboxed assistant)? Send the user the upload_page link to drop the file in a browser, then continue with the file_url.

Example: remove vocals

Just ask your assistant in plain language:

You: Remove the vocals from https://example.com/song.mp3

Assistant calls:  separate_audio { "url": "...", "task": "vocals" }
            ->  { "job_id": "01H...", "status": "in_process_queue" }

Assistant calls:  check_audio_job { "job_id": "01H..." }
            ->  { "done": true, "outputs": [
                    { "track": "vocals",        "download_url": "https://..." },
                    { "track": "instrumentals", "download_url": "https://..." }
                ] }

The free tools work the same way — for example "convert this file to wav" calls convert_audio and returns a download URL instantly, with no credits used.

Things to try

• ›Remove the vocals from https://example.com/song.mp3 and give me the karaoke track.
• ›Split this track into stems — vocals, drums, bass, guitar and piano.
• ›Convert https://example.com/interview.m4a to MP3 at 128 kbps.
• ›Extract just the audio from this video: https://example.com/clip.mp4
• ›Denoise this podcast and remove the room echo: https://example.com/episode.wav
• ›Make an 8D version and trim it to 30 seconds.
• ›What is the duration and bitrate of https://example.com/track.flac?
• ›Check my credit balance.

Limits & notes

• Input files must be public http(s) URLs, up to 100 MB each.
• AI separation runs asynchronously — always poll check_audio_job for the result.
• Credits are charged per minute of audio, only for AI separation.
• The free audio tools never use credits.

Privacy & security

• Your audio is fetched from the URL you pass, processed, and deleted — we do not store your files.
• Your API token is scoped to your account; revoke it any time from the API page.
• The endpoint is HTTPS only; we block URLs that resolve to private/internal addresses.

Troubleshooting

Back to the MCP server overview