ElevenLabs Runtime Skill
SkillMediaGenerate audio through the ElevenLabs MCP. Use when the user asks to create a voiceover / text-to-speech, multi-speaker dialogue, a sound effect or ambience, or music.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the ElevenLabs Runtime Skill skill
What this skill tells your AI
The instructions your AI receives, as published by manor-os/manor-ai in packages/core/ai/skills/mcp_elevenlabs/SKILL.md and read by ahel’s review.
Use this skill to generate audio via ElevenLabs through the ElevenLabs MCP (mcp__elevenlabs__*).
When To Use
Use ElevenLabs when the user asks for a spoken voiceover, a multi-speaker dialogue, a sound effect/ambience, or generated music. For images/video use mcp_replicate or mcp_jimeng.
Connection
Authenticates with an ElevenLabs API key. On an auth error, stop and ask the user to fix the key. Generation consumes credits — treat each call as billable.
Core Tools
list_voices— the user's prebuilt + cloned voices (pick avoicebefore TTS).text_to_speech— text → MP3 voiceover.text_to_dialogue— a list of speaker lines → one natural multi-speaker track.generate_sound_effect— a dedicated SFX/ambience bed.compose_music— generate music/score audio.
Common Recipes
Voiceover
list_voices→ choose avoice(confirm with the user if they have a preference). 2. Confirm the script. 3.text_to_speech.
Two-person dialogue
list_voicesfor the speakers. 2. Build the ordered speaker lines. 3.text_to_dialogue.
Guardrails
- Each generation consumes credits — confirm the script/voice before generating; don't regenerate repeatedly without asking.
- Voice cloning / likeness: only use voices the user is entitled to; don't synthesize a real person's voice without their consent.
- Keep within reasonable length per call; very long scripts burn credits fast — chunk deliberately.
- Refuse disallowed content (impersonation, fraud, etc.).
Edge Cases & Errors
- Pick a valid
voiceid fromlist_voices— a wrong/empty voice fails or uses an unintended default. - Output is an audio asset reference — return it; don't claim success before it's produced.
- Auth/quota errors → stop and tell the user (key invalid or credits exhausted).
Signals
- GitHub stars
- 171
- Forks
- 52
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
mcp-elevenlabs- Source
- github.com/manor-os/manor-ai