Let an AI agent process audio
Connect a remote Model Context Protocol server instead of writing REST integration code. The agent can import audio, start a job, monitor it, and return a finished file with explicit tool calls.
Connect a compatible client
Add a remote HTTP MCP server in the client settings. The client discovers OAuth automatically and opens Neural Analog in a browser. Sign in, review the requesting client, and approve or deny the connection.
Server URL
https://api.neuralanalog.com/mcp- 1
Add the URL
Choose HTTP or streamable HTTP if asked.
- 2
Authorize in the browser
Sign in to Neural Analog and approve the named client.
- 3
Return to the client
The client completes the OAuth exchange and loads the tools.
Available tools
The first release keeps the surface small and maps each tool to one visible product action.
Import
import_audio_from_url- Import one supported public track URL. Use the REST API for playlists and creator pages.
prepare_audio_upload + complete_audio_upload- Send local file bytes directly, or use a secure browser upload handoff to select the file.
Process
restore_audio- Repair or upscale a track or stem.
master_audio- Master a track without running a hidden restoration.
split_stems- Separate a track with the requested stem preset.
transcribe_to_midi- Convert a track or stem into MIDI.
Monitor and deliver
get_job_status- Check an import or processing job.
get_account_usage- Read the current plan and processing balance.
get_download_link- Create a short-lived link for a completed artifact.
Treat audio work as a background job
Import and processing tools return immediately. The result is a stable handle, not the finished audio.
Every job-creating tool requires an idempotency_key. Generate one value for a logical request and reuse that same value for every retry, including after reconnecting the client. Use a new value when you intentionally want a new job.
- 1
Keep the handle
Save object_type, object_id, audio_id, state, poll_after_seconds, and the idempotency key used to create it.
- 2
Poll once per interval
Call get_job_status after poll_after_seconds and keep polling the same active job.
- 3
Stop on a terminal state
Download only after is_complete. If is_failed is true, stop and use error_message.
Upload a local file safely
Start with prepare_audio_upload. Upload the file bytes through the returned upload URL, or open the secure browser upload link and select the file there.
Direct byte upload
Call prepare_audio_upload, PUT the bytes to the returned upload URL or multipart URLs, then call complete_audio_upload.
Browser upload
Open the returned browser_upload_url. After sign-in, the page verifies that the current account owns the reserved audio ID and uploads the selected file directly. Return to the client after the confirmation appears.
Access, processing minutes, and disconnection
MCP uses the same account, plan limits, and processing balance as the web Studio and REST API. Read-only status and usage checks do not start processing. Importing a file does not run an audio model.
MCP uses an OAuth connected-app grant. Remove the server from the client to stop using it, then revoke its grant in Settings to invalidate that client's refresh tokens and sessions without affecting other integrations.