MCP 服务器

MCP Voice Soundboard

A text-to-speech MCP server with 48 voices across 9 languages, supporting emotion spans, SFX tags, and multi-speaker dialogue. Deployable via a single npx command with built-in guardrails and swappable backends.

README

48 voices • 9 languages • 5 presets • 8 emotions • SSML-lite • SFX tags • multi-speaker dialogue Swappable TTS backends. Guardrails built in. Ships as a single <code>npx</code> command.

Highlights

MCP native — stdio transport, works with Claude Desktop, Cursor, and any MCP client
5 tools — voice_speak, voice_dialogue, voice_status, voice_interrupt, voice_inner_monologue
48 approved voices, 9 languages — English (American + British), Japanese, Mandarin, Spanish, French, Hindi, Italian, Brazilian Portuguese. Curated presets: narrator, announcer, whisper, storyteller, assistant
Emotion spans — 8 emotions via {joy}...{/joy} inline markup
SSML-lite — <break>, <emphasis>, <prosody> without full SSML complexity
SFX tags — [ding], [chime], [whoosh], [tada], [pop], [click] inline sound effects
Multi-speaker dialogue — Speaker: line format with auto-cast and pause directives
Guardrails — rate limiting, concurrency semaphore, request timeouts, path traversal protection, secret redaction
Swappable backends — Mock (built-in), HTTP proxy, Python bridge, or bring your own

Quick Start

npx @mcptoolshop/voice-soundboard-mcp

Or install globally:

npm install -g @mcptoolshop/voice-soundboard-mcp
voice-soundboard-mcp

Claude Desktop / MCP Client Config

Add to your MCP client configuration (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "voice-soundboard": {
      "command": "npx",
      "args": ["-y", "@mcptoolshop/voice-soundboard-mcp"]
    }
  }
}

With options:

{
  "mcpServers": {
    "voice-soundboard": {
      "command": "npx",
      "args": [
        "-y", "@mcptoolshop/voice-soundboard-mcp",
        "--artifact=path",
        "--output-dir=/tmp/voice-output",
        "--timeout=30000",
        "--max-concurrent=2"
      ]
    }
  }
}

MCP Tools

`voice_speak`

Synthesize speech from text.

text:         "Hello world!"
voice?:       "am_fenrir"          # Voice ID or preset name
speed?:       1.0                  # 0.5 - 2.0
mood?:        "dry"                # Humor mood (sensor-humor integration)
format?:      "wav"                # wav | mp3 | ogg | raw
artifactMode?: "path"             # path | base64
outputDir?:   "subdir"            # Subdirectory within output root
sfx?:         true                # Enable [ding], [chime] etc.

`voice_dialogue`

Multi-speaker dialogue synthesis.

script:       "Alice: Hello!\nBob: Hey there!"
cast?:        { "Alice": "af_sky", "Bob": "am_fenrir" }
speed?:       1.0
concat?:      true                 # Combine into single file
debug?:       true                 # Include cue_sheet
artifactMode?: "path"             # path | base64
outputDir?:   "subdir"            # Subdirectory within output root

`voice_status`

Returns engine health, available voices, presets, and backend info. No arguments.

`voice_interrupt`

Stop or rollback active synthesis.

streamId?:    "stream-123"
reason?:      "user_spoke"         # user_spoke | context_change | timeout | manual

`voice_inner_monologue`

Ephemeral micro-utterances for ambient narration. Requires --ambient flag or VOICE_SOUNDBOARD_AMBIENT_ENABLED=1.

text:         "Interesting..."     # Max 500 chars, auto-redacted
category?:    "thinking"           # general | thinking | observation | debug

Voices

48 voices across 9 languages. Language is auto-inferred from the voice ID prefix — no configuration required.

Prefix	Language
`af_` / `am_`	English (American)
`bf_` / `bm_`	English (British)
`jf_` / `jm_`	Japanese
`zf_` / `zm_`	Mandarin Chinese
`ef_` / `em_`	Spanish
`ff_`	French
`hf_` / `hm_`	Hindi
`if_` / `im_`	Italian
`pf_` / `pm_`	Brazilian Portuguese

English — American

ID	Name	Gender	Style
`af_aoede`	Aoede	Female	Musical
`af_bella`	Bella	Female	Warm
`af_heart`	Heart	Female	Caring
`af_jessica`	Jessica	Female	Professional
`af_kore`	Kore	Female	Youthful
`af_nicole`	Nicole	Female	Soft
`af_sarah`	Sarah	Female	Clear
`af_sky`	Sky	Female	Airy
`am_eric`	Eric	Male	Confident
`am_fenrir`	Fenrir	Male	Powerful
`am_liam`	Liam	Male	Friendly
`am_michael`	Michael	Male	Deep
`am_onyx`	Onyx	Male	Smooth
`am_puck`	Puck	Male	Playful

English — British

ID	Name	Gender	Style
`bf_alice`	Alice	Female	Proper
`bf_emma`	Emma	Female	Refined
`bf_isabella`	Isabella	Female	Warm
`bm_fable`	Fable	Male	Storytelling
`bm_george`	George	Male	Authoritative
`bm_lewis`	Lewis	Male	Friendly

Japanese

ID	Name	Gender	Style
`jf_alpha`	Alpha	Female	Clear
`jf_gongitsune`	Gongitsune	Female	Storytelling
`jf_nezuko`	Nezuko	Female	Gentle
`jf_tebukuro`	Tebukuro	Female	Warm
`jm_kumo`	Kumo	Male	Calm

Mandarin Chinese

ID	Name	Gender	Style
`zf_xiaobei`	Xiaobei	Female	Bright
`zf_xiaoni`	Xiaoni	Female	Gentle
`zf_xiaoxiao`	Xiaoxiao	Female	Clear
`zf_xiaoyi`	Xiaoyi	Female	Warm
`zm_yunjian`	Yunjian	Male	Authoritative
`zm_yunxi`	Yunxi	Male	Friendly
`zm_yunxia`	Yunxia	Male	Calm
`zm_yunyang`	Yunyang	Male	Confident

Spanish

ID	Name	Gender	Style
`ef_dora`	Dora	Female	Warm
`em_alex`	Alex	Male	Confident
`em_santa`	Santa	Male	Jolly

French

ID	Name	Gender	Style
`ff_siwis`	Siwis	Female	Refined

Hindi

ID	Name	Gender	Style
`hf_alpha`	Alpha	Female	Clear
`hf_beta`	Beta	Female	Warm
`hm_omega`	Omega	Male	Deep
`hm_psi`	Psi	Male	Calm

Italian

ID	Name	Gender	Style
`if_sara`	Sara	Female	Warm
`im_nicola`	Nicola	Male	Confident

Brazilian Portuguese

ID	Name	Gender	Style
`pf_dora`	Dora	Female	Warm
`pm_alex`	Alex	Male	Confident
`pm_santa`	Santa	Male	Jolly

Presets

Preset	Voice	Speed	Description
`narrator`	`bm_george`	0.95	Calm, clear, documentary style
`announcer`	`am_eric`	1.1	Bold, energetic, broadcast style
`whisper`	`af_sky`	0.85	Soft, intimate, gentle
`storyteller`	`bf_emma`	0.90	Expressive, varied pacing
`assistant`	`af_jessica`	1.0	Friendly, helpful, conversational

Emotion Spans

Wrap text in emotion tags to control prosody and voice routing:

{joy}Great news!{/joy} But {calm}let me explain.{/calm}

Supported: neutral, serious, friendly, professional, calm, joy, urgent, whisper

CLI Flags

Flag	Default	Description
`--artifact=path\|base64`	`path`	Audio delivery mode
`--output-dir=<path>`	`<tmpdir>/voice-soundboard/`	Output directory
`--backend=mock\|http\|python`	`mock`	Backend selection
`--ambient`	off	Enable inner-monologue system
`--max-concurrent=<n>`	`3`	Max concurrent synthesis requests
`--timeout=<ms>`	`60000`	Per-request timeout
`--retention-minutes=<n>`	`240`	Auto-cleanup age (0 to disable)

Packages

This is a pnpm monorepo with two publishable packages:

Package	Description	npm
`@mcptoolshop/voice-soundboard-core`	Backend-agnostic core library (validation, SSML, chunking, schemas)
`@mcptoolshop/voice-soundboard-mcp`	MCP server with CLI, guardrails, and transport

Development

# Install
pnpm install

# Build
pnpm build

# Test (363 tests)
pnpm test

# Build + test in one step
pnpm verify

Part of MCP Tool Shop

Project Structure

mcp-voice-soundboard/
  packages/
    core/               @mcptoolshop/voice-soundboard-core
      src/
        limits.ts         SHIP_LIMITS, text/chunk limits
        schemas.ts        VoiceRequest, VoiceResponse, error codes
        artifact.ts       resolveOutputDir, path sandbox
        voices.ts         Approved voice registry + presets
        emotion.ts        Emotion span parser
        ssml/             SSML-lite parser + limits
        chunking/         Text chunker
        sfx/              SFX tag parser + registry
        sandbox.ts        Safe filenames, symlink checks
        ambient.ts        AmbientEmitter for inner monologue
        redact.ts         PII/secret redaction
    mcp-server/         @mcptoolshop/voice-soundboard-mcp
      src/
        server.ts         MCP tool registration + guardrail wiring
        cli.ts            CLI entrypoint (stdio transport)
        backend.ts        Backend abstraction + mock/HTTP
        concurrency.ts    SynthesisSemaphore
        rateLimit.ts      ToolRateLimiter (sliding window)
        timeout.ts        withTimeout utility
        retention.ts      Output file cleanup timer
        redact.ts         Server-level redaction
        validation.ts     Synthesis result validation
        tools/            Individual tool handlers
  assets/               Logo, audio event manifests
  docs/                 Architecture docs

Privacy

No telemetry. This tool collects no usage data, sends no analytics, and makes no network requests except to the TTS backend you configure. All processing is local.

Security

See SECURITY.md for vulnerability reporting.

See THREAT_MODEL.md for the full threat surface analysis.

Project	Description
soundboard-plugin	Claude Code plugin — slash commands, emotion-aware narration

Support

Questions / help: Discussions
Bug reports: Issues
Security: SECURITY.md

Scorecard

Category	Score	Notes
A. Security	10/10	SECURITY.md, THREAT_MODEL.md, redaction, no telemetry
B. Error Handling	8/10	Structured error contract (code/hint/retryable), toToolError pattern
C. Operator Docs	9/10	README, CHANGELOG, HANDBOOK, tool docs
D. Shipping Hygiene	9/10	CI, verify script, dependabot, lockfile
E. Identity	10/10	Logo, translations, landing page, metadata
Total	46/50

License

MIT

Built by <a href="https://mcp-tool-shop.github.io/">MCP Tool Shop</a>