Brainiall TTS MCP Server

Hosted text-to-speech for AI agents via the Model Context Protocol.

54 neural voices, 9 languages — including native Brazilian Portuguese (pf_dora, pm_alex, pm_santa) — served from a hosted, pay-per-use API. No GPU, no model downloads, no ElevenLabs subscription: bring one API key and pay $0.008 per 1,000 characters ($10 free credits on signup).

Why this server

Every other TTS MCP server either runs models locally (heavy, slow to set up) or wraps a third-party key you already pay a subscription for. This one is a hosted, metered API:

• Zero setup — remote server, nothing to install
• Pay per use — $0.008/1K characters, billed against your Brainiall balance
• $10 free credits — sign up at app.brainiall.com
• WAV out — 16-bit PCM, 24 kHz mono, returned as playable MCP audio content or base64 JSON

Quick start

Get an API key at app.brainiall.com ($10 welcome credits, no card required).

VS Code / GitHub Copilot

The button installs the remote server in discovery mode, so VS Code can list its tools without a secret. Synthesis still fails closed until you add your Brainiall API key. For a secure workspace configuration that prompts once and stores the key in VS Code's secret storage, copy .vscode/mcp.json into your project or clone this repository, then start brainiallTts from MCP: List Servers.

{
  "inputs": [
    {
      "type": "promptString",
      "id": "brainiall-api-key",
      "description": "Brainiall API key",
      "password": true
    }
  ],
  "servers": {
    "brainiallTts": {
      "type": "http",
      "url": "https://api.brainiall.com/mcp/tts/mcp",
      "headers": {
        "Authorization": "Bearer ${input:brainiall-api-key}"
      }
    }
  }
}

After the server starts, try:

Use Brainiall TTS to list the Brazilian Portuguese voices, then read “Olá do VS Code” with pf_dora.

The install URL format and secret-input configuration follow the official VS Code MCP guide and configuration reference.

GitHub Copilot cloud agent

Use the restricted remote-server configuration for Copilot cloud agent or Copilot code review. It allowlists the three Brainiall tools and resolves the Bearer token from the repository's COPILOT_MCP_BRAINIALL_API_KEY Agents secret; no API key is committed.

Continue (VS Code / JetBrains)

Copy the ready-to-use Continue MCP block to .continue/mcpServers/brainiall-tts.yaml, then put BRAINIALL_API_KEY=your-key in .continue/.env. Keep that .env file out of version control.

The block uses Continue's supported streamable-http transport and resolves the key through ${{ secrets.BRAINIALL_API_KEY }}; no credential is committed. See the Continue setup and smoke test.

Dify builders

For a Dify speech workflow, use the open-source BRAINIALL Speech provider and keep the API key in Dify's credential store. The bounded Dify intent route explains the caller-owned boundary; it is not a Dify partnership or a guarantee of production behavior.

Remote server (recommended — nothing to install)

Codex CLI / IDE / ChatGPT desktop app

Use the project-scoped Codex configuration. It reads BRAINIALL_API_KEY from the environment, allowlists the three Brainiall tools, and prompts before every call.

Claude Code

Use the environment-backed Claude Code configuration. It keeps the key out of .mcp.json, scopes the server to the current project, and documents the external-data and metered-usage boundary.

Claude Desktop / any client with .mcp.json-style config

{
  "mcpServers": {
    "brainiall-tts": {
      "type": "http",
      "url": "https://api.brainiall.com/mcp/tts/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_BRAINIALL_API_KEY"
      }
    }
  }
}

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "brainiall-tts": {
      "url": "https://api.brainiall.com/mcp/tts/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_BRAINIALL_API_KEY"
      }
    }
  }
}

LM Studio

Use the official one-click deeplink and the explicit local-model privacy boundary in the LM Studio setup guide.

OpenCode v2

Use the remote-server schema and environment-backed Bearer header in the OpenCode v2 setup guide.

Msty Studio

Use the manual Msty Studio field guide for a Streamable HTTP connection. The reference file is not presented as an official integration or import preset.

Hugging Face Chat UI (self-hosted)

Use the administrator MCP_SERVERS example and a tool-capable model. The guide documents the shared-key boundary for base servers.

Clients that require /v1/audio/speech

Use the zero-dependency partial OpenAI-shaped TTS adapter when a client can configure an OpenAI-style speech route but cannot call Brainiall's native endpoint. The adapter returns WAV only, declares the sample rate observed in the WAV header, never retries a metered request, and documents the exact compatibility limits. It is a local reference adapter, not a claim that the hosted Brainiall API is natively OpenAI-compatible.

Run locally (stdio-free, still calls the hosted API)

The server itself is a thin wrapper — you can self-host it and it will proxy to api.brainiall.com with your key:

docker build -t brainiall-tts-mcp .
docker run -p 8080:8080 -e APIM_KEY=YOUR_BRAINIALL_API_KEY brainiall-tts-mcp
# MCP endpoint: http://localhost:8080/mcp

Tools

Tool	Description	Cost
synthesize_speech	Convert text (≤5000 chars) to WAV speech. Params: text, language (default pt), voice, speed (0.5–2.0), output_format (audio | base64_json)	$0.008/1K chars
list_voices	Full voice catalog with language, gender, accent, quality grade. Optional language filter	free
check_tts_service	Backend health status	free

Voices

Language	Voices
Portuguese (BR)	pf_dora, pm_alex, pm_santa
English (US)	af_heart, af_bella, af_nova, am_adam, am_michael + 14 more
English (GB)	bf_alice, bf_emma, bm_daniel, bm_george + 4 more
Spanish	ef_dora, em_alex, em_santa
French	ff_siwis, fm_gilles
Italian	if_sara, im_nicola
Hindi	hf_alpha, hf_beta, hm_omega, hm_psi
Japanese	jf_alpha, jf_gongitsune, jm_kumo + 2 more
Mandarin	zf_xiaoxiao, zm_yunxi + 6 more

Example

Ask your agent:

"Read this paragraph out loud in Brazilian Portuguese with a female voice"

The agent calls synthesize_speech(text=..., language="pt", voice="pf_dora") and receives playable WAV audio.

For copy-ready prompts and smoke tests for narration, accessibility, language practice, and agent alerts, see Agent workflow recipes.

For REST client testing and automation, import the Postman collection. It contains health, voice-list and two-character synthesis smoke tests and keeps the API key in a collection variable rather than the request URL.

For no-code automation, import the n8n text-to-WAV workflow. It uses n8n's Header Auth credential instead of embedding a key in the workflow, sends editable text, voice and speed fields to the hosted API, and returns the WAV in the binary property speech. The template is statically validated; select your credential and run it in your own n8n instance to validate the live integration.

For code generation, API clients, and directory discovery, use the machine-readable OpenAPI 3.1 specification. The specification documents the authenticated voice catalog and WAV synthesis endpoints without embedding an API key.

Authentication & billing

Pass your Brainiall API key as a Bearer token (see configs above). Usage is metered per character against your account balance — the same key works across all Brainiall APIs (STT, OCR, NLP, image and more at brainiall.com).

Endpoints

• MCP (Streamable HTTP): https://api.brainiall.com/mcp/tts/mcp
• Health: https://api.brainiall.com/mcp/tts/health
• Underlying REST API: POST https://api.brainiall.com/v1/tts/synthesize

License

MIT