ElevenLabs

ElevenLabs

This connector works with ElevenLabs (AI voice: text to speech, speech to text, dubbing and voice agents) through the ElevenLabs API at https://api.elevenlabs.io with the user's own API key: voices and the public Voice Library, models, speech generation with a download link, transcription of audio and video, the history of generated audio, plan and credit usage, dubbing projects, and the voice agents with their conversations and transcripts.

Spending: speech, transcription and dubbing are billed to the user's ElevenLabs account (credits on a subscription plan, US dollars on pay-as-you-go API billing). Everything else is free. Before a call that costs money, tell the user what it will roughly cost and wait for a yes. elevenlabs_get_subscription shows the credits left.

Setup (about 3 minutes)

1. Sign in at https://elevenlabs.io and open your API keys (https://elevenlabs.io/app/settings/api-keys). For a key that does not depend on one person, a workspace admin can create it for a service account under Workspace settings, Service Accounts.
2. Create a key and either leave it unrestricted or allow at least these endpoints: Text to Speech and Flows (speech), Speech to Text, Models (read), Voices (read), Add voice from Voice Library, Speech History (read; write only for deletes), User (read), Dubbing (read and write), and ElevenAgents (read). A credit quota on the key caps what it can spend; an IP allowlist would block AnythingMCP.
3. Copy the key (it starts with sk_ and is shown only once), paste it into ELEVENLABS_API_KEY and install the connector. It is sent as the xi-api-key header, only to ElevenLabs.

Data residency: accounts on an isolated environment use their own host and their own keys: https://api.eu.residency.elevenlabs.io (EU), https://api.in.residency.elevenlabs.io (India) or https://api.sg.residency.elevenlabs.io (Singapore). Change the connector's base URL to that host; a key from an isolated environment is refused on api.elevenlabs.io with 401, and the other way round.

How to work

1. Voices: elevenlabs_list_voices for the account's voices; elevenlabs_get_voice for one. elevenlabs_search_voice_library searches the public Voice Library; elevenlabs_add_library_voice saves a Library voice to the account (uses a voice slot) so it can speak.
2. Speech: elevenlabs_create_speech with the text and a voice id starts the audio and returns an id. Then call elevenlabs_get_speech with that id: after a few seconds status is completed and content_url is an MP3 link the user can open or download (it expires after about an hour; call elevenlabs_get_speech again for a fresh link). elevenlabs_list_speech_generations finds earlier ones. Models: eleven_multilingual_v2 (default), eleven_v3 (most expressive, audio tags such as [whispers]), eleven_flash_v2_5 (fastest, half the price). elevenlabs_list_models lists all models with their limits.
3. Transcription: elevenlabs_transcribe with source_url (a public file link, YouTube or TikTok) returns the text right away; diarize: true separates speakers, and elevenlabs_get_transcript then shows who said what and when. Use file_url only when ElevenLabs cannot open the link (AnythingMCP then uploads the file, up to 10 MB).
4. History: elevenlabs_list_history and elevenlabs_get_history_item show what was generated (text, voice, date). elevenlabs_delete_history_item is installed switched off.
5. Account: elevenlabs_get_subscription for plan, credits used and left, reset date and voice slots.
6. Dubbing: elevenlabs_create_dubbing with a link and a target language starts a dub and returns dubbing_id and the expected processing time. Check elevenlabs_get_dubbing every 30 to 60 seconds until status is dubbed; elevenlabs_list_dubbings lists all. elevenlabs_get_dubbing_transcript returns the subtitles (SRT or WebVTT). The dubbed video or audio is watched and downloaded in the ElevenLabs app under Dubbing.
7. Voice agents: elevenlabs_list_agents and elevenlabs_get_agent (prompt, first message, voice, LLM). elevenlabs_list_conversations lists calls with outcome and title (summary_mode: include adds summaries); elevenlabs_get_conversation returns the full transcript and analysis.

Costs (October 2026): speech about 1 credit per character with eleven_multilingual_v2 and eleven_v3, 0.5 with eleven_flash_v2_5 (pay-as-you-go API: $0.08 or $0.04 per 1,000 characters); transcription about $0.22 per hour of audio; dubbing about $0.33 a minute with a watermark, $0.50 without. Failed speech generations are not charged.

Audio files: the connector returns audio as download links, never as raw audio in the chat. Speech comes from elevenlabs_get_speech as content_url. Older history items, dubs and agent call recordings are played and downloaded in the ElevenLabs app.

Common errors (the reason is in detail.code and detail.message)

• 401 invalid_api_key or unauthorized: the key is wrong, expired or disabled, or belongs to an isolated environment (change the base URL). A key that is cut off or mistyped is refused with 400 invalid_api_key.
• 402 insufficient_credits: no credits left this period, or the key's own credit quota is used up.
• 403 insufficient_permissions: the key is not allowed to use this endpoint, or an IP allowlist blocks the request; edit the key.
• 400 or 422: a parameter the model does not accept (for example language_code with eleven_multilingual_v2, or a text above the model's limit); detail.param names it.
• 404 voice_not_found and similar: check the id with the list tools.
• 429 rate_limit_exceeded, concurrent_limit_exceeded or system_busy: wait a moment and retry once. Do not retry in a loop.

Out of scope here: sound effects and music (the API returns only raw audio and keeps no history of them), real-time and streaming audio, voice cloning and voice design, speech to speech, audio isolation, Studio projects, editing or creating agents, phone calls and batch calls, and downloading audio files of history items, dubs or agent calls.