Voice Mcp

io.github.forgemeshlabs由 GitHub 账号 forgemeshlabs 发布v0.2.7更新于 Nov 3, 2026

Give Your Agent A Voice: x402 pay-per-call speech with 20 voices, 10 personas, 31 languages.

已验证STDIO仅桌面媒体与设计AI 与机器学习

概览

AI 生成的概览

本地 MCP 服务器,通过按次付费的 x402 语音 API 把文本转为 WAV 语音,提供 20 种声音、10 种人格和 31 种语言。

可以做什么
  • 列出语音目录:声音、人格、语言、价格档位、字符上限与语速/质量范围
  • 用 31 种语言的标准声音把文本合成为语音
  • 以细粒度语速和质量控制生成语音
  • 生成 Storyteller、Narrator、Urgent 等富有表现力的人格语音
  • 按 OpenAI 风格的请求(输入、声音、模型、响应格式)生成语音
  • 一次调用为最多 20 段文本批量生成音频
适用场景
适合需要让助手开口说话的场景:旁白、状态播报、告警、品牌或角色声音、演示,以及脚本化或批量音频序列。建议先用免费的目录工具挑选声音并估算费用。
运行要求
以 stdio 方式在本地运行,通过 npm 包 @forgemeshlabs/voice-mcp 安装(需 Node.js),也可用 Docker。需要一个持有 USDC 的 Base 主网钱包私钥,通过 WALLET_PRIVATE_KEY 提供,因为每次生成调用都要付费。可选的 X402_MAX_PRICE_USD 和 X402_SESSION_BUDGET_USD 只能下调支出上限。需要访问固定的后端地址。
安装前请注意
服务器会用 WALLET_PRIVATE_KEY 中的私钥在 Base 上签署 USDC 付款,请使用专用的低余额钱包。每次生成调用都会花钱(约每次 0.005 至 0.01 美元),免费的目录工具不收费。提交的文本会发送给第三方语音服务。内置上限会限制收款方、网络、资产和单次金额,可选预算变量只能进一步下调。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Voice Mcp,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

Voice MCP

[M8ven Verified]

[npm version] [npm downloads] [License: MIT] [Node.js] [payments] [network] [mcpservers.org]

Give Your Agent A Voice: x402 pay-per-call speech with 20 voices, 10 personas, 31 languages, granular speed and quality controls, OpenAI-shaped requests, and batch audio.

This MCP wraps https://voice.forgemesh.io, an x402 Voice API with standard voices, persona voices, OpenAI-shaped speech requests, 31 languages, speed controls, quality controls, and batch generation. Payments are made per call in USDC on Base.

Voice Coverage

  • 10 standard voices: M1-M5, F1-F5
  • 10 persona voices: Storyteller, Narrator, Announcer, Assistant, Urgent, Sage, Spark, Anchor, Velvet, Echo
  • 31 languages: en, ko, ja, ar, bg, cs, da, de, el, es, et, fi, fr, hi, hr, hu, id, it, lt, lv, nl, pl, pt, ro, ru, sk, sl, sv, tr, uk, vi
  • Granular control: speed 0.7x-2.0x, quality steps 1-100, persona selection, OpenAI-shaped audio format requests, and batch generation for up to 20 texts
  • Voice samples are generated on demand by the paid speech tools and returned as audio_base64 WAV output

Voice Samples

Tools

ToolPricePurpose
list_voice_catalogFreeVoices, personas, languages, pricing, buckets, and controls
generate_standard_voice$0.005Low-cost speech with 10 standard voices
generate_controlled_voice$0.005 / $0.006Speech with granular speed and quality controls
generate_persona_voice$0.005 / $0.01Storyteller, Velvet, Narrator, Announcer, Assistant, Urgent, and more
generate_openai_compatible_voice$0.005OpenAI-shaped /v1/audio/speech request
generate_batch_voices$0.005Up to 20 texts per call

Short prices apply to 1-500 characters. Long prices apply to 501-2000 characters.

The MCP validates voice names, language codes, audio formats, speed/quality ranges, batch item count, and character limits locally before making a paid x402 call.

Install

bash
npm install -g @forgemeshlabs/voice-mcp

Docker

Build:

bash
docker build -t voice-mcp .

Run over stdio:

bash
docker run --rm -i \  -e WALLET_PRIVATE_KEY=0x... \  voice-mcp

CMD arguments:

json
["node", "index.js"]

MCP Config

json
{  "mcpServers": {    "voice": {      "command": "voice-mcp",      "env": {        "WALLET_PRIVATE_KEY": "0x..."      }    }  }}

Requirements and safety

  • Use a dedicated, low-balance wallet; the server signs payments with WALLET_PRIVATE_KEY.
  • The server refuses to sign for any payee other than the ForgeMesh Voice wallet, any network other than Base mainnet, any asset other than USDC, or any amount over the built-in per-call cap ($0.01).
  • Optional env X402_MAX_PRICE_USD (per call) and X402_SESSION_BUDGET_USD (cumulative per process, default $10) can only LOWER the built-in caps.
  • The backend URL is fixed to https://voice.forgemesh.io; all requests are same-origin, time-limited (60s), size-capped (2 MB) and never follow redirects.

Notes

  • Paid tools require a Base wallet private key with USDC.
  • The server returns audio_base64 for audio tools so MCP clients can store, play, or forward the WAV bytes.
  • No API keys or subscriptions are required for the voice service itself.

来源:README.md,提交 a5f423b

工具

6

list_voice_catalog

List Voice Catalog

Free discovery tool. Lists all 20 voices, 10 persona voices, 31 language codes, price buckets, character limits, and granular speed/quality controls before a paid voice generation call.

无需输入参数。

generate_standard_voice

Generate Standard Voice

Generate low-cost WAV speech from text using one of 10 standard voices across 31 languages. Best for simple agent narration, status updates, alerts, and short spoken responses. Costs $0.005 for 501-2000 chars.

输入参数langtext必填voice

generate_controlled_voice

Generate Controlled Voice

Generate WAV speech with granular controls for speed and quality. Use this when an agent needs faster, slower, clearer, more polished, or more deliberate delivery. Costs $0.005 for 1-500 chars or $0.006 for 501-2000 chars.

输入参数langtext必填speedstepsvoice

generate_persona_voice

Generate Persona Voice

Generate expressive WAV speech with persona voices such as Storyteller, Narrator, Announcer, Assistant, Urgent, Sage, Spark, Anchor, Velvet, or Echo. Best for branded agents, characters, demos, stories, alerts, and premium user experiences. Costs $0.005 for 1-500 chars or $0.01 for 501-2000 chars.

输入参数langtext必填speedstepsvoice

generate_openai_compatible_voice

Generate OpenAI-Compatible Voice

Generate speech using an OpenAI-shaped request with input, voice, model, and response_format fields. Use this for agents or apps already designed around /v1/audio/speech style payloads. Costs $0.005 for 501-2000 chars.

输入参数input必填modelvoiceresponse_format

generate_batch_voices

Generate Batch Voices

Generate WAV audio for up to 20 text items in one paid call using standard voices. Best for queues, notifications, scripted sequences, content batches, and multi-step agent workflows. Costs $0.005 for 501-2000 total chars.

输入参数items必填defaults

版本历史

4
  1. v0.2.7最新Nov 3, 2026
  2. v0.2.6Nov 1, 2026
  3. v0.2.5Oct 8, 2026
  4. v0.2.2Sep 16, 2026

评分

已满足适用检查项的 90%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分
  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分