Clone a voice from an audio sample. Creates a new voice with a Cartesia-assigned ID; the voice is available only within your workspace.

Create a new pronunciation dictionary in your workspace.

Delete a pronunciation dictionary from your workspace.

Delete a cloned voice from your workspace.

Download a file from cloud storage (e.g. output of `text_to_speech` with `save=true`) to the MCP server. Optionally refresh the download link.

Retrieve credit usage and balance for your Cartesia account (admin API key required).

List pronunciation dictionaries in your workspace.

List available voices.

Transcribe speech audio. Accepts WAV, MP3, OGG, FLAC, AIFF, or raw PCM. Returns transcript and optional word-level timestamps.

Generate speech audio from text. By default (`save=true`) the audio is persisted in Cartesia cloud storage and the response includes `file_id` and `download_url` (24-hour public link). Hosted clients (Claude, ChatGPT) should use `download_url`. `file_path` is a copy on the MCP server host — useful for local `uvx` and for server-side tools like `speech_to_text` in the same MCP session.

Update a pronunciation dictionary by adding or replacing items.

Apply a different voice to existing audio. The original audio is transcribed (optionally using a provided transcript), and then re-spoken in the target voice.