Developers
Give your AI assistant a voice.
Polyvox is an MCP server. Meta Muse, Claude Code, Cursor, or any agent that speaks the Model Context Protocol can make clips in your cloned voice, in 75 languages, and hand you a link to share.
Connect in three steps
- 1. Create a key. Sign in, open Settings → API keys, and create one. It is shown once.
- 2. Add the endpoint to your client with the key as a bearer token.
- 3. Ask for a clip. “Say ‘happy birthday, Ana’ in Spanish in my voice and give me the link.”
MCP endpoint
https://api.usepolyvox.com/mcp
Streamable HTTP, stateless. Auth: OAuth 2.1 sign-in for clients that support it, or Authorization: Bearer pvx_…
Meta Muse
Tell Muse: “Connect Polyvox. It’s an MCP server at https://api.usepolyvox.com/mcpand needs a bearer token.” When Muse asks for credentials, paste your key into its secure credential prompt, not into the chat.
Claude Code
claude mcp add --transport http polyvox https://api.usepolyvox.com/mcp \
--header "Authorization: Bearer pvx_YOUR_KEY"Cursor and other JSON-configured clients
{
"mcpServers": {
"polyvox": {
"url": "https://api.usepolyvox.com/mcp",
"headers": { "Authorization": "Bearer pvx_YOUR_KEY" }
}
}
}ChatGPT
No key needed. Add Polyvox as a custom connector with the endpoint above and OAuth authentication. The first time ChatGPT needs your account it asks you to sign in to Polyvox and approve the connection. You can disconnect it any time in Settings → API keys.
Tools
| Tool | What it does | Cost |
|---|---|---|
| get_account | Plan, clips left this month, per-clip text limit, languages included. | Free |
| list_voices | Your cloned voices plus the ready-made voice actors, with the languages each can speak. | Free |
| list_languages | All 75 languages with codes and which ones your plan includes. | Free |
| generate_clip | Text in, spoken MP3 out. Translates into the target language first. Returns a share link and a one-hour audio URL. | 1 credit |
| get_clip | Status, share link, fresh audio URL and play count for one clip. | Free |
| list_clips | Your most recent saved clips. | Free |
| suggest_script | Draft a short script from a description, to review before speaking it. | Free |
| clone_voice | Clone the user's voice from a short recording (URL or base64). Requires the user's consent and the language spoken. | 1 voice slot |
| get_clone_link | A one-tap, signed-in link to the clone studio so the user records in the browser. Works once, 15 minutes. | Free |
REST API
The same key works against the REST API that powers the web app. The OpenAPI document is at https://api.usepolyvox.com/openapi.json, so clients that build integrations from an API spec (Muse included) can use it directly.
Each example lists your voices, makes a clip with the first one, and saves the MP3. Copy one and swap in your key.
curl
# 1. Pick a voice (copy an "id")
curl https://api.usepolyvox.com/api/v1/voices \
-H "Authorization: Bearer pvx_YOUR_KEY"
# 2. Make a clip (waits until the audio is ready, up to ~3 minutes)
curl https://api.usepolyvox.com/api/v1/generations --max-time 300 \
-H "Authorization: Bearer pvx_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"voice_id":"VOICE_ID","transcript":"Hello!","language":"es"}'
# 3. Download the MP3 (use the "id" from step 2)
curl "https://api.usepolyvox.com/api/v1/generations/GENERATION_ID/audio?download=true&redirect=false" \
-H "Authorization: Bearer pvx_YOUR_KEY" -o clip.mp3Python
import requests
API = "https://api.usepolyvox.com/api/v1"
HEADERS = {"Authorization": "Bearer pvx_YOUR_KEY"}
voice = requests.get(f"{API}/voices", headers=HEADERS).json()[0]
clip = requests.post(
f"{API}/generations",
headers=HEADERS,
json={"voice_id": voice["id"], "transcript": "Hello!", "language": "es"},
timeout=300,
)
clip.raise_for_status()
clip_id = clip.json()["id"]
audio = requests.get(
f"{API}/generations/{clip_id}/audio",
headers=HEADERS,
params={"download": "true", "redirect": "false"},
timeout=120,
)
audio.raise_for_status()
open("clip.mp3", "wb").write(audio.content)PowerShell
$Api = "https://api.usepolyvox.com/api/v1"
$Headers = @{ Authorization = "Bearer pvx_YOUR_KEY" }
$Voice = (Invoke-RestMethod "$Api/voices" -Headers $Headers)[0]
$Body = @{ voice_id = $Voice.id; transcript = "Hello!"; language = "es" } | ConvertTo-Json
$Clip = Invoke-RestMethod "$Api/generations" -Method Post -Headers $Headers `
-ContentType "application/json; charset=utf-8" -Body $Body -TimeoutSec 300
Invoke-WebRequest "$Api/generations/$($Clip.id)/audio?download=true&redirect=false" `
-Headers $Headers -OutFile clip.mp3- The request returns once the audio is ready. A long script can take up to about 3 minutes, so give your client a timeout of at least 180 seconds.
- Retrying after a timeout is safe and free. While the clip is still being made, the retry gets
409 generation_in_progresswith itsgeneration_id; once it is done, the same request returns that clip without spending another credit. - Download the audio from
GET /api/v1/generations/{id}/audio?download=true&redirect=false. Withoutredirect=falseit answers with a redirect to a one-hour storage link instead.
Limits, security and privacy
- Every clip spends one credit, the same meter as the web studio. Paid plans: 5,000 characters per clip, up to all 75 languages.
- A new account without a plan gets one free demo clip through its key: up to 200 characters, downloadable, so you can hear the voice and run your integration end to end. After that, clips answer
402with aurlto the plans. - 120 requests per minute per key. Up to 5 active keys per account.
- A key can only see your own voices, make clips with them, and clone new ones within your plan’s slots (at most 3 clones per day per key, and only with your consent recorded). It cannot change billing or your profile, or manage keys: those need a signed-in browser session.
- Keys are stored hashed. Revoke one at any time from Settings and it stops working immediately.
- Clips made by an agent appear in My Clips like any other. Share links are public to anyone who has the URL, as always.
- See the privacy policy and terms. Questions: contact us.