TTS API · MCP Server

Text in. Studio-quality audio out.

Production-ready MP3 narration through a simple REST API—or straight from Claude, Cursor, and any MCP agent. 69 languages, your own cloned voices, and word-level timestamps included.

From $6 per million characters · pay as you go

Read the docs

No SDK to install. No monthly fee. Credit never expires. Demand is high right now, so every request is reviewed—most hear back within a business day.

curl

# 1. Create a speech job
curl -s -X POST "https://api.everlit.audio/v1/speech" \
  -H "Authorization: Bearer api_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello from Everlit.", "voice": "669"}'
# => 202 { "id": "spch_01jx...", "status": "queued",
#          "audio_url": "https://api.everlit.audio/v1/speech/spch_01jx.../audio" }

# 2. Poll until "succeeded" (typically 15-60s), then download.
curl -sL -o hello.mp3 -H "Authorization: Bearer api_YOUR_KEY" \
  "https://api.everlit.audio/v1/speech/spch_01jx.../audio"

per million characters: $6
languages: 69
typical job time: 15–60s
MCP tools: 8

Two ways in

Call it from code, or let your agent call it

The same API key works for both. Send text, get back a job with a stable audio URL, and pick it up when it’s done—by polling, by webhook, or by asking your assistant.

REST API

Any language with an HTTP client

speak.py

import os, time, requests

BASE = "https://api.everlit.audio"
H = {"Authorization": f"Bearer {os.environ['EVERLIT_API_KEY']}"}

job = requests.post(f"{BASE}/v1/speech", headers=H,
                    json={"text": "Hello from Everlit.", "voice": "669"}).json()

while job["status"] in ("queued", "processing"):
    time.sleep(3)
    job = requests.get(f"{BASE}/v1/speech/{job['id']}", headers=H).json()

with open("hello.mp3", "wb") as f:
    f.write(requests.get(job["audio_url"], headers=H).content)

MCP server

Claude, Cursor, Claude Code, and any MCP client

mcp.json

{
  "mcpServers": {
    "everlit-tts": {
      "type": "http",
      "url": "https://api.everlit.audio/v1/mcp",
      "headers": { "Authorization": "Bearer api_YOUR_KEY" }
    }
  }
}

Then just ask: “Narrate this article in Penny’s voice and give me the MP3.” The agent picks a voice, submits the job, waits politely, and hands back the link.

AI-forward by design

Give your agent a voice

Everlit is a first-class MCP server. Every REST endpoint is exposed as a tool, so an assistant can browse voices, narrate a draft, check on the job, and pull word timings without you writing a line of glue code.

Same key, same bill, same limits
Tools run the exact code the REST routes run. Nothing is metered or validated differently because an agent asked.
Polite polling, built in
Running jobs return how many seconds to wait. The server tells agents not to spin, so a chatty model can't burn your read budget.
Errors an agent can act on
Tool errors carry the same structured envelope as the REST API, plus the HTTP status and a retry-after hint on 429s.
Streamable HTTP, no OAuth dance
Stateless transport with a bearer header. Paste one config block and you're connected.

8 tools

POST https://api.everlit.audio/v1/mcp

create_speech

Submit text and a voice; get a job back in milliseconds.

get_speech

Check a job. Running jobs tell the agent how long to wait.

list_speech

Browse recent jobs, filter by status, search by text.

get_speech_timings

Word-level timestamps for a finished job.

delete_speech

Remove audio before its 30-day expiry.

list_voices

Everlit's catalog plus your own cloned voices.

list_languages

All 69 languages with their billing weights.

get_usage

Characters, requests, and spend for any date range.

Pricing

One number. No plan required.

Prepaid and independent of any Everlit subscription. Add a card, buy credit, and every successful job is deducted at the published rate—in exact fractions of a cent.

Pay as you go

$6/ million characters

For Latin, Cyrillic, and Greek scripts. Other scripts are weighted so every language costs about the same per finished hour of audio—see below.

Word-level timestamps included
Cloned voices at the same rate
Signed webhooks and idempotency keys
Audio hosted for 30 days on a stable URL
Failed jobs are never charged
Credit never expires

Top up from $5 to $10,000 once approved. Already a customer? Add credit in Studio.

What that buys

English text, ×1 script weight

A 1,200-word article
~7,000 chars
$0.04
A 10,000-word long read
~57,000 chars
$0.34
A 90,000-word audiobook
~515,000 chars
$3.09
1,000 articles a month
~7M chars
$42

Script-weighted characters

English speaks about 1,050 characters a minute; Chinese about 300. Weights keep the price per hour of audio level across languages.

×1Latin, Cyrillic & Greek scripts
$6.00 / M
×1.5Arabic, Hebrew & Indic scripts
$9.00 / M
×2Hangul & Thai
$12.00 / M
×2.5Kana (Japanese)
$15.00 / M
×3Han (Chinese)
$18.00 / M

Counted with standard Unicode script ranges, so your bill is reproducible from your own request logs. The optional language hint can never raise it, and weights only ever change in your favor.

60 / min

speech jobs created

300 / min

reads, polls & MCP messages

concurrent jobs

20,000

characters per request

Safety rails, not credits. Need more headroom? Limits are set per account—email support@everlit.audio.

Languages

69 languages, every official EU language included

Word-level timestamps ship today in 25 of them, and coverage keeps expanding. Everything is published live on the languages endpoint.

Word timings availableAudio only, for now

Latin, Cyrillic & Greek scripts ×1

47 languages

English (word timings available)
Spanish (word timings available)
Portuguese (word timings available)
French (word timings available)
German (word timings available)
Italian (word timings available)
Dutch (word timings available)
Polish (word timings available)
Russian (word timings available)
Ukrainian (word timings available)
Czech (word timings available)
Slovak (word timings available)
Romanian (word timings available)
Greek (word timings available)
Hungarian (word timings available)
Finnish (word timings available)
Swedish (word timings available)
Danish (word timings available)
Bulgarian (word timings available)
Croatian (word timings available)
Estonian (word timings available)
Latvian (word timings available)
Lithuanian (word timings available)
Maltese (word timings available)
Slovenian (word timings available)
Irish
Norwegian
Catalan
Galician
Turkish
Vietnamese
Indonesian
Serbian
Bosnian
Belarusian
Kazakh
Mongolian
Uzbek
Georgian
Icelandic
Welsh
Basque
Esperanto
Swahili
Kinyarwanda
Ganda
Kabyle

Arabic, Hebrew & Indic scripts ×1.5

17 languages

Arabic
Hebrew
Persian
Urdu
Uighur
Kurdish (Sorani)
Bengali
Hindi
Tamil
Telugu
Kannada
Marathi
Gujarati
Nepali
Malayalam
Punjabi
Odia

Hangul & Thai ×2

2 languages

Korean
Thai

Kana (Japanese) ×2.5

1 language

Japanese

Han (Chinese) ×3

2 languages

Chinese
Cantonese

Word-level timestamps

Your text, verbatim, with a clock on every word

Every finished job can return the words you submitted—punctuation and all—each with a start and end time. They come from your text, not from a speech recognizer’s best guess, so “$5” is one entry whose span covers the spoken “five dollars.”

Karaoke-style read-along highlighting
Captions and subtitles that match the script exactly
Seek to any sentence in a long recording
Included with every job, at no extra cost

GET /v1/speech/{id}/timings

{
  "object": "speech.timings",
  "id": "spch_01jx...",
  "duration_seconds": 58.3,
  "words": [
    { "word": "Hello",    "start": 0.00, "end": 0.31 },
    { "word": "from",     "start": 0.38, "end": 0.55 },
    { "word": "Everlit.", "start": 0.61, "end": 1.12 }
  ]
}

Built for production

The boring parts, done properly

Everything you’d expect from an API you’re going to put in a nightly pipeline.

Voice cloning

Clone your narrators or brand voice in Studio. Cloned voices show up in the catalog and bill at the same rate as ours.

Word timings, free

Precise per-word start and end times in 25 languages, taken from your submitted text rather than a transcript guess.

69 languages

One catalog spanning Latin, Cyrillic, Greek, Arabic, Hebrew, Indic, Hangul, Thai, kana, and Han scripts. Mixed-script text just works.

Signed webhooks

Pass a callback URL and we POST the finished job, HMAC-signed, with retries for 30 minutes. No polling loop required.

Stable audio URLs

The download URL is known the instant you submit and stays valid for 30 days. Store it once; no expiring links to refresh.

Idempotency keys

Retry a create safely. The same key and body returns the original job with no second charge, for 30 days.

Hash-only API keys

Keys are shown once and stored as a SHA-256 digest. Revoke instantly from Studio; every response carries rate-limit headers.

MCP native

Eight tools over Streamable HTTP with structured errors and polite polling hints, so agents behave in production.

OpenAPI spec

A published OpenAPI 3 document and a Markdown mirror of the docs, ready for code generators and AI assistants alike.

Asynchronous on purpose

Jobs finish in about 15 to 60 seconds because every file gets a full render and an automatic quality check. That makes Everlit ideal for articles, archives, audiobooks, courses, and publish-time pipelines—and the wrong choice for live conversational agents or IVR that need speech in milliseconds. If your app can wait half a minute, you get richer audio for less.

Getting started

Three steps to your first MP3

Demand for the API is high right now, so we review every request to keep capacity where our customers expect it. Request a key, and once you’re approved everything else happens in Studio.

Request access

Tell us what you're building and roughly how much text you'll send. With current demand we review every request by hand, usually within a business day.

Add credit, mint a key

Once approved, open Billing → TTS API in Studio, add a card and starting credit from $5, then create an api_ key under API → Keys.

Send text

POST to /v1/speech, or paste the MCP config into Claude or Cursor and ask for a narration. Usage shows up in Studio within minutes.

Already approved? Open Studio

Ship audio this afternoon.

One endpoint, one MCP config, $6 per million characters. Read the reference, request a key, and your first narration is a curl away.

Read the API docs

What you’ll get

Studio-quality MP3 narration, 44.1 kHz
69 languages and your own cloned voices
Word-level timestamps in 25 languages, free
REST API plus an 8-tool MCP server
Webhooks, idempotency, 30-day stable URLs
Pay as you go from $5, at $6 per million characters