# https://mcp.koe.live — Koe（声）

Koe is a voice service: clone-voice TTS (濱田優貴 / Yuki Hamada's voice), STT,
and a voice messenger. Everything below works WITHOUT an account
(5 calls/day per IP); a free email key raises the limit to 200/day.

Voice messenger: https://mcp.koe.live/app   (browser PWA — record → send to any email)
MCP endpoint:    https://mcp.koe.live/mcp   (Streamable HTTP, JSON-RPC 2.0)
OpenAPI:         https://mcp.koe.live/openapi.json
This file:       https://mcp.koe.live/llms.txt

## Try it right now (no auth, 1 curl)

Send a clone-voice message — the recipient gets an email with a ▶listen URL,
no app/install needed on their side:

  curl -X POST https://mcp.koe.live/api/messages/send \
    -H 'content-type: application/json' \
    -d '{"to":"friend@example.com","from_name":"MyAgent","text":"こんにちは、Koeからの声です"}'
  → {"ok":true,"id":"…","listen_url":"https://mcp.koe.live/v/<file>","emailed":true}

Three send modes (`mode`):
  tts     — text → clone voice            {"text":"…"}                (default when text given)
  raw     — recorded audio sent as-is     {"audio_base64":"…","mime":"audio/webm"}
  revoice — recording → transcript (scribe_v1) → clone voice (eleven_v3)
            {"audio_base64":"…","mime":"…","mode":"revoice"} → response includes "transcript"

The recipient can REPLY from the listen page without an account
(POST /api/messages/{id}/reply — same body, no `to`; it returns to the sender).

## Auth & limits

- Anonymous: 5 calls/day per IP; text ≤120 chars.
- Free key:  POST https://mcp.koe.live/api/auth/request-link {"email":"you@…"} → click the
  emailed link → your `koe_…` key. 200/day, text ≤500 chars.
  Send it as `Authorization: Bearer <key>`. With a key you also get an inbox.
- OAuth 2.1 (PKCE S256) for MCP clients:
  discovery at https://mcp.koe.live/.well-known/oauth-authorization-server

## REST endpoints

POST /api/messages/send          send a voice message (see above)
POST /api/messages/call          Bearer — invite {to} to a LIVE voice call;
                                 creates a koe.live WebRTC room, emails them a
                                 join link, returns room_url for the caller
GET  /api/messages/inbox         Bearer — your received messages (max 100)
POST /api/messages/{id}/read     Bearer — mark read
POST /api/messages/{id}/reply    no auth — reply to the original sender
POST /api/auth/request-link      get a key by email ({"redirect":"app"} opens /app signed-in)
GET  /audio/{file}               audio bytes (mp3/webm/m4a)
GET  /v/{file}                   human listen page (+ no-login reply recorder)

## MCP

  claude mcp add --transport http koe https://mcp.koe.live/mcp

Tools:
  list_voices — available voices
  speak       — text → clone-voice MP3 (returned inline as audio content)
  send_voice  — voice message to an email (text / raw audio / revoice)
  inbox       — your received voice messages (key required)
  transcribe  — audio → text (scribe_v1; audio_base64 or audio_url)

Voice: 濱田優貴 clone (Japanese). Operated by Yuki Hamada / Enabler Inc.