Cartesia (Voice AI) MCP. Synthesize and clone any voice in real-time.

You're connecting your AI client to Cartesia Voice AI. This server gives you high-performance voice synthesis and speech recognition tools. It lets your agent generate lifelike audio and transcribe any spoken file, all while giving you deep control over the voice profile itself.

Synthesizing Speech from Text (Text-to-Speech)
When you need text to sound like a human talking, you use these tools. You can call tts_bytes to get raw audio bytes based on any text input. If your application needs audio immediately—like in real time—you'll use tts_sse. This streams synthesized audio data using Server-Sent Events (SSE), so it plays out as the model generates it.

The server handles advanced models, giving you high fidelity and low latency output.

Transcribing Audio to Text (Speech-to-Text)
If you send an audio file, this server returns a text transcript. You can process large amounts of material at once using stt_batch, which handles transcribing entire batches of files into text format across multiple languages. This makes bulk processing straightforward.

Voice Cloning and Modification
Need a specific voice? You use clone_voice to create a custom voice model. Just five seconds of audio is all it takes; the server replicates that speaker's unique tone and pitch. If you already have an audio clip but need to change how it sounds, call voice_changer_bytes.

This modifies the vocal characteristics while making sure it keeps the original speaker’s natural intonation intact. You can also use infill_bytes to generate audio data that smoothly connects two separate audio segments together.

Managing Voice Profiles and Agents
You've got a bunch of voices, right? To see what you have, call list_voices, which lists every available model, whether it’s standard or one you cloned. You can also use list_agents to see every voice agent set up in the system. If you need details on a specific profile, you'll get them using get_voice (for metadata) or get_agent (for detailed instance info).

When your voice needs an update—maybe its name changed or some parameters shifted—you use update_voice. To clean house and remove a model permanently, call delete_voice. Similarly, you can modify agent settings with update_agent or get historical data by calling list_agent_calls, which shows all calls and transcripts for a given agent.

Localization and Pronunciation Control
Sometimes the voice needs to speak in an accent or language it wasn't trained on. You use localize_voice to adapt an existing voice profile so it can speak fluently in a new dialect or language. For specialized terminology, you manage pronunciation dictionaries. To start fresh, call create_pronunciation_dict and save a custom dictionary that guides how the AI pronounces specific words or terms.

You can then modify this saved data using update_pronunciation_dict, or wipe it out entirely with delete_pronunciation_dict. If you want to check what dictionaries are already set up, use list_pronunciation_dicts.

Utility and Workflow Management
To keep your workflow running smoothly, you'll first need a temporary access token; you get that using generate_access_token for protected client-side requests. You can always check how much juice you've got left by calling get_usage_credits, which shows current credits and when the next billing cycle refreshes. To keep things tidy, if you want to see every single pronunciation dictionary saved, you use list_pronunciation_dicts.