VoiceMode MCP — 使用場景, 安裝 & 即時演示

為什麼要用

核心特性

Local Whisper option — no cloud audio
Multiple TTS backends: OpenAI, ElevenLabs, local Coqui
Push-to-talk or voice-activated modes
Streams partial responses so you hear Claude thinking
Works in terminal alongside Claude Code CLI

即時演示

實際使用效果

voicemode-mcp.replay ▶ 就緒

0/0

安裝

選擇你的客戶端

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "voicemode-mcp": {
      "command": "uvx",
      "args": [
        "voice-mode"
      ]
    }
  }
}

開啟 Claude Desktop → Settings → Developer → Edit Config。儲存後重啟應用。

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "voicemode-mcp": {
      "command": "uvx",
      "args": [
        "voice-mode"
      ]
    }
  }
}

Cursor 使用與 Claude Desktop 相同的 mcpServers 格式。專案級設定優先於全域。

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "voicemode-mcp": {
      "command": "uvx",
      "args": [
        "voice-mode"
      ]
    }
  }
}

點擊 Cline 側欄中的 MCP Servers 圖示，然後選 "Edit Configuration"。

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "voicemode-mcp": {
      "command": "uvx",
      "args": [
        "voice-mode"
      ]
    }
  }
}

格式與 Claude Desktop 相同。重啟 Windsurf 生效。

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "voicemode-mcp",
      "command": "uvx",
      "args": [
        "voice-mode"
      ]
    }
  ]
}

Continue 使用伺服器物件陣列，而非映射。

~/.config/zed/settings.json

{
  "context_servers": {
    "voicemode-mcp": {
      "command": {
        "path": "uvx",
        "args": [
          "voice-mode"
        ]
      }
    }
  }
}

加入 context_servers。Zed 儲存後熱重載。

claude mcp add voicemode-mcp -- uvx voice-mode

一行命令搞定。用 claude mcp list 驗證，claude mcp remove 移除。

使用場景

實戰用法： VoiceMode

Drive a Claude Code session hands-free while reading on another screen

👤 Devs who read docs or designs on one monitor while coding ⏱ ~30 min intermediate

何時使用： You're reading a design doc and want to dictate changes without alt-tabbing.

前置條件

Microphone + speakers — System audio configured — test with say "hello" or equivalent
Whisper model ready — voice-mode install-whisper downloads the local model

步驟

Start voice

Use voicemode. Listen for prompts and speak responses. Repeat after me: "ready"✓ 已複製

→ TTS plays "ready"
Dictate a change

[spoken] Update src/auth.ts — use bcrypt instead of plain SHA256 for passwords.✓ 已複製

→ Transcription correct; change applied; TTS confirms
Review

[spoken] Read me the diff.✓ 已複製

→ TTS reads diff in chunks, pausable

結果： A working session where your hands never leave what they were doing.

注意事項

TTS talking over your prompts — Enable push-to-talk mode or a wake word

搭配使用： filesystem

Code by voice for accessibility or RSI recovery

👤 Devs with RSI, low vision, or preferring speech input ⏱ ~60 min intermediate

何時使用： You can't type for a while and need to keep shipping.

前置條件

Tolerable ambient noise — Quiet room; headset mic beats laptop mic

步驟

Baseline

[spoken] Use voicemode. Read the latest git diff out loud, pausing between files.✓ 已複製

→ Clear TTS read
Workflow

[spoken] Refactor the user model in src/models/user.ts. Move password hashing into a method. Show me the plan first.✓ 已複製

→ Plan spoken; confirmation required before changes

結果： A full coding session without keyboard input.

注意事項

Code symbols mispronounced by TTS — Configure the TTS phoneme dictionary for common programming terms

組合

與其他 MCP 搭配，撬動十倍槓桿

voicemode-mcp + filesystem

Voice-dictated code changes land in the repo

I'll dictate changes; apply them in files after reading each back.✓ 已複製

voicemode-mcp + github

Dictate a PR description after voice-reviewing the diff

Read me the staged changes, then open a PR with a description I'll dictate.✓ 已複製

工具

此 MCP 暴露的能力

工具	輸入參數	何時呼叫	成本
start_listening	mode: "ptt"\|"vad"	Begin a voice session	free or OpenAI Whisper API
speak	text: str, voice?: str	Any time Claude wants to surface something audibly	TTS provider-dependent
transcribe_last	none	Fetch what the user just said	Whisper call
stop_listening	none	End voice session	free

成本與限制

運行它的成本

API 配額: Local: free. OpenAI Whisper: $0.006/min. ElevenLabs TTS: ~$0.30/1k chars.
每次呼叫 Token 數: Audio pipelines are not token-costed directly
費用: Free with local stack; metered with cloud providers
提示: Local Whisper + Coqui TTS is totally free but lower quality — start cloud, downgrade later

安全

權限、密鑰、影響範圍

最小權限： microphone speakers

憑證儲存： TTS/STT API keys in env

資料出站： Voice audio to TTS/STT provider if not local

Never use cloud STT in calls with confidential audio unless you trust the provider's retention policy

故障排查

常見錯誤與修復

Mic not detected

System audio permission — grant terminal/Claude Code mic access

驗證： `voice-mode test-mic` prints levels

TTS sounds robotic

Default is Coqui local — switch to OpenAI tts-1-hd via VOICE_MODE_TTS=openai

Lag between my speech and response

Use local Whisper-tiny for STT; cloud adds 500ms+

替代方案

VoiceMode 對比其他方案

替代方案	何時用它替代	權衡
macOS Dictation + say command	You just want basic OS-level voice	No integration with Claude's output — one-way only
Superwhisper / Wispr Flow	You want a polished native macOS dictation app	Not MCP-integrated; no agent-level workflows

VoiceMode

為什麼要用

核心特性

即時演示

實際使用效果

安裝

選擇你的客戶端

使用場景

實戰用法： VoiceMode

Drive a Claude Code session hands-free while reading on another screen

前置條件

步驟

注意事項

Code by voice for accessibility or RSI recovery

前置條件

步驟

注意事項

組合

與其他 MCP 搭配，撬動十倍槓桿

工具

此 MCP 暴露的能力

成本與限制

運行它的成本

安全

權限、密鑰、影響範圍

故障排查

常見錯誤與修復

替代方案

VoiceMode 對比其他方案

更多

資源