convez.ai
MCP ServerIntegração & APIsIntermediário
workblock100

Edge TTS MCP

workblock100/edge-tts-mcp

Text-to-speech neural gratuito com áudio inline no Claude

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que integra síntese de voz neural gratuita do Microsoft Edge ao Claude, com áudio reproduzido inline no chat. Oferece duas ferramentas principais: geração de áudio com controle de voz, taxa e tom, e listagem de vozes disponíveis por idioma. Usa a biblioteca open-source edge-tts, sem custos ou chaves de API necessárias.

Instalação

$npx convez-ai workblock100/edge-tts-mcp

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerar áudio de respostas do Claude para acessibilidade ou consumo multimídia
  • Criar conteúdo em múltiplos idiomas com vozes naturais
  • Prototipar aplicações que requerem síntese de fala sem custos

Veredito

Ferramenta bem construída que aproveita APIs públicas do Microsoft para entregar TTS de qualidade sem custo. Ideal para quem quer adicionar áudio natural ao Claude com setup simplificado; requer gerenciamento manual da URL do túnel Cloudflare após cada restart.

README

A tradução para o português deste README ainda está na fila.

Edge TTS MCP

Free, high-quality neural text-to-speech as a Claude custom connector. Audio plays inline in the chat through an MCP App iframe — no money, no API keys.

Uses Microsoft's neural voices via the open-source edge-tts library.

What it gives you

Two tools claude.ai will see:

  • speak(text, voice?, rate?, pitch?) — generates audio and renders an inline player widget right in the conversation. The audio autoplays (when the browser allows it) and the player shows the voice, the source text, and standard play controls.
  • list_voices(language?) — lists every neural voice for a locale prefix (en, en-GB, es-MX, etc.).

Recommended voices: en-US-AriaNeural (default, warm female), en-US-GuyNeural (male), en-US-JennyNeural (friendly female), en-GB-SoniaNeural (British female), en-GB-RyanNeural (British male).

One-time setup

cd ~/Documents/edge-tts-mcp
./bootstrap.sh   # writes .env with a fresh MCP_SECRET, installs deps

Run

./start.sh   # starts uvicorn server + cloudflared tunnel, prints the connector URL
./stop.sh    # stops both

start.sh rotates the PUBLIC_URL in .env automatically. The cloudflared URL changes every restart, so you'll need to update the connector URL in claude.ai each time the tunnel comes back up. Logs at /tmp/edge-tts-mcp.log and /tmp/edge-tts-cf.log.

Add to claude.ai

  1. Open https://claude.ai/settings/connectors.
  2. Click Add custom connector.
  3. Paste the Claude connector URL printed by start.sh (looks like https://<random>.trycloudflare.com/<secret>/mcp).
  4. Authentication: None.
  5. Save. The next time you open a chat, the speak and list_voices tools will be available.

How the inline player works

Tool calls return base64-encoded MP3 in _meta.audio_base64. The speak tool declares _meta.ui.resourceUri = "ui://edge-tts/player.html" — an MCP App resource served by this server. claude.ai renders that HTML in a sandboxed iframe inside the chat, the iframe receives the tool result via the MCP App postMessage protocol, decodes the base64 to a Blob, sets it on an <audio controls> element, and tries to autoplay.

Troubleshooting

  • No tools showing up in Claude — re-check the connector URL includes /<MCP_SECRET>/mcp at the end. Without the secret path you'll get a 404.
  • Tunnel URL keeps changing — that's how Cloudflare quick tunnels work (free, account-less). Either accept it and re-paste the URL after each start.sh run, or set up a named tunnel via cloudflared tunnel login.
  • "Failed to initialize" in the player — usually means the iframe couldn't load https://esm.sh/@modelcontextprotocol/ext-apps. Check the iframe console via the host's devtools.
  • Audio doesn't autoplay — browser autoplay policy. The <audio controls> element shows a play button you can click.

💼 Hire the author

Built by Elijah — I build custom MCP servers, Python automations, web scrapers, and AI chatbots. Fixed-price from $85, working sample before you pay.

🛒 Ready-made products: MCP Server Starter Kit · MCP Pro Pack · all 6 →

Ver o original no GitHub

Bom para

Relacionados