Menu
Text-to-speech with Mexican Spanish voices (ElevenLabs) and voice cloning. OpenAI TTS fallback.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Add the EasyBits MCP server (@easybits.cloud/mcp) so the container agent can manage cloud files, images, video/voice generation, designs, websites, forms, brand kits, characters, and a built-in DB. 65+ tools (server-side).
Reference & priority guide for the EasyBits MCP server (`mcp__easybits__*`). When to use EasyBits vs native skills, and which EasyBits tools exist.
Semantic search over your workspace text files using local multilingual embeddings (fastembed + e5-small). Free, local.
Extract text from images locally with PaddleOCR weights via onnxruntime. Free, multi-language, batch.
Remove backgrounds and segment images locally with the u2net family (rembg). Free, no API keys, works in batch.
Generate vector logos (SVG) and convert raster images to SVG via Recraft on fal.ai. Use for logo / brand mark / icon requests, or when the user sends a PNG/JPG and wants the editable SVG version.
| name | voice |
| description | Text-to-speech with Mexican Spanish voices (ElevenLabs) and voice cloning. OpenAI TTS fallback. |
| allowed-tools | Bash(text-to-speech:*),Bash(clone-voice:*),Bash(ffmpeg:*),Bash(yt-dlp:*) |
Generate audio when the user asks for voice or when replying to a voice note feels natural.
| Voice | Style | When to use |
|---|---|---|
antonio | Confident, gentle, latino | Default — everyday conversation |
jc | Energetic, broadcaster | News, data, exciting updates |
brian | Warm, soft, podcast | Long explanations, calm tone |
daniel | Young, natural, casual | Casual banter, young audience |
enrique | Rich, credible, narrator | Serious narration, formal reports |
maya | Dynamic, Mexican female | Female storytelling, energetic |
cristina | Young, conversational, Mexican female | Female casual chat |
regina | Sweet, friendly, Mexican female | Female warm/professional |
custom | Cloned voice | When the group has a cloned voice |
text-to-speech "Qué onda, aquí el resumen de hoy" antonio
text-to-speech "Última hora: el servidor está al 99% de uptime" jc
text-to-speech "Te explico cómo funciona el sistema de pagos" brian
text-to-speech "Esto con la voz personalizada" custom
mcp__nanoclaw__send_message({ text: "voice", audio_path: "/workspace/agent/tts-XXX.ogg" })
clone-voice /workspace/agent/attachments/audio-123.ogg "voice-name"
ffmpeg -i /workspace/agent/attachments/video.mp4 -vn -acodec libopus /workspace/agent/extracted-audio.ogg -y
clone-voice /workspace/agent/extracted-audio.ogg "voice-name"
YouTube blocks all requests without cookies. The container needs /workspace/youtube-cookies.txt mounted RO. If it's not present, tell the user to register the mount via additionalMounts in the group's container.json.
cp /workspace/youtube-cookies.txt /tmp/yt-cookies.txt
yt-dlp --cookies /tmp/yt-cookies.txt -x --audio-format wav -o "/workspace/agent/yt-audio.%(ext)s" "YOUTUBE_URL"
clone-voice /workspace/agent/yt-audio.wav "voice-name"
The cookies file must be copied to /tmp/ first because yt-dlp needs to write to it.
Warn the user first — YouTube + cloning takes 1-2 minutes:
mcp__nanoclaw__send_message({ text: "Descargando audio de YouTube y clonando la voz, dame 1-2 min..." })
For best results, pick a video with clear speech (no music, no background noise).
The script saves to /workspace/agent/voice_config.json and you can use text-to-speech "text" custom thereafter. To stop using the cloned voice: delete that file.
401 Unauthorized from ElevenLabs / OpenAI: OneCLI secret not assigned. Tell the user to re-run /add-voice on the host.yt-dlp: command not found: package missing from the agent's container. Re-run /add-voice on the host (it adds yt-dlp to packages.apt).