unlimitedtts

How to Give Your AI Agent a Voice With a Text-to-Speech API

By Chris Graney-Ward ·

AI agents are good at producing text, but plenty of jobs end in audio: a spoken summary, a narrated report or a voice message. A text-to-speech API gives an agent a way to finish those jobs on its own. It sends text and gets an audio file back.

The Unlimited TTS API is built for this. It is a paid service designed for AI assistants and automated systems, and an agent can use it without a person creating an account first. If you just want to listen to something yourself, the free text-to-speech reader is the better fit.

Connect your assistant

There are three routes, depending on what your assistant supports:

  • MCP: assistants that support local MCP servers can add the Unlimited TTS server and get ready-made tools for listing voices, getting a price and creating audio.
  • OpenClaw: the plugin bundles the MCP server with an Unlimited TTS agent skill.
  • Plain HTTP: any assistant with web or HTTP tools can read the live agent guide and API contract, then make the requests itself. The API page has a prompt you can copy.

Choose how the agent pays

Each option is priced from the same per-character rate, so the choice comes down to how the agent works:

  • Pay per request over x402: suits an agent with its own wallet that only makes occasional requests. There is no account or balance to set up, but every request waits for payment to settle.
  • Prepaid x402 balance: suits sustained traffic. You deposit once, and each request is then authorized by a wallet signature instead of a separate payment.
  • Card credits: suits a person or app paying by card through Stripe. This is the only option with asynchronous jobs and stored audio, and the published MCP and OpenClaw integrations do not use it yet.

Keep spending under your control

Letting software spend money deserves some care. A sensible setup looks like this:

  1. Start with a quote. Have the agent show you the voice, exact price and payment method before it creates any audio.
  2. Set a budget. The MCP server and OpenClaw plugin support per-request and daily spending limits.
  3. Approve the first payment yourself. Check that it matches what you expected before letting the agent carry on.
  4. Keep your keys out of the chat. An external wallet signer handles payments, so never paste a private key or seed phrase into a conversation.

Crypto charges are final, so an agent should also stop and ask you if a payment's outcome is unknown rather than paying again. Before funding a wallet, check the live payment details, and never send funds to an address copied from an unverified prompt.

What happens to your text

Requests paid over x402 are synchronous and do not enter the storage workflow. If an agent should not reveal its text before paying, it can get a quote using only the text's length and a hash. Asynchronous card-credit audio is kept by default so it can be downloaded later, and you can opt out of that for each request.

Get started

Models, voices, limits and prices can change, so your agent should read them from the live contract rather than rely on a blog post. The API for AI Agents page has setup instructions and links to the current details, and the integrations are on GitHub.