ElevenLabs Text to Speech API + MCP vs Inworld Realtime TTS-2

ElevenLabs Text to Speech API + MCP

Keys can be limited to chosen endpoints and given a credit quota, and service accounts hold keys that don't belong to a person. The Terms let ElevenLabs train on content unless you opt out, which the Eleven v4 launch page contradicts, and v4 runs through Text to Dialogue rather than the classic text-to-speech endpoint.

Inworld Realtime TTS-2

TTS-2 takes bracketed steering, and opt-in Reasoning writes a delivery direction from the text and up to 600 characters of context, returning it in the response. Per-character prices are public and the free plan needs no card. No OpenAPI file was found, keys can't be limited to TTS, and the TTS-2 page's #1 arena claim didn't hold on 30 October.

ElevenLabs Text to Speech API + MCP

1. Call Eleven v4 through POST /v1/text-to-dialogue and set model_id to eleven_v4, because that endpoint defaults to eleven_v3.
2. Retrain an older Instant or Professional Voice Clone on v4 before using it there, since earlier clones aren't tuned for the new model.
3. Spell out numbers yourself on eleven_flash_v2_5, which doesn't normalise them by default, and turning that on is Enterprise only.
4. On 429 read code, back off on rate_limit_exceeded and wait for running requests on concurrent_limit_exceeded.
5. Give the agent a key scoped to Text to Speech with a credit quota and, for a fixed server, an IP allowlist.

Inworld Realtime TTS-2

1. Send the copied Base64 credential as-is in Authorization: Basic. Don't encode it again.
2. Set modelId explicitly, because SDK defaults may name deprecated models.
3. Keep text under 2,000 UTF-16 units on /tts/v1/voice and 4,000 on voice:stream, or use an async job.
4. Turn on Reasoning with synthesisContext.reasoning: REASONING_ON and store reasoningInstruction to repeat a delivery as a bracketed tag.
5. On 429 or error code 8, back off exponentially with jitter. No Retry-After header is documented.