Skip to main content
Glama
Sign In
VoxFabric

VoxFabric

by YottaMeta
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

声络 VoxFabric

Weave Every AI Voice

MCP speech I/O for any agent host — local mic/speakers, Edge TTS speak, optional local ASR listen.

VoxFabric is a stdio MCP server: any MCP-capable host can call tools to speak and listen on the machine that runs this process. Speak uses Edge TTS (network to Microsoft/Bing). Listen is optional and can use local ASR (e.g. Whisper) with models you place yourself. There is no web UI — MCP only. The project is source-available; personal/non-commercial evaluation is fine, but commercial use and 二开 require paid written authorization (see LICENSE).

Tools

Tool

Role

speak

TTS playback (Edge)

listen

Mic → text

listen_start / get_speech / listen_stop

Persistent listen

set_engine / catalog / health

Engines & health

Optional engines include local Whisper / SenseVoice / Aliyun ASR, energy or Silero VAD, and Edge TTS.

Related MCP server: VOICEVOX TTS MCP

Install & connect

All prerequisites, Windows / Kali steps, MCP JSON, proxy, and troubleshooting: DEPLOY.md.

License

Source-available — not OSI “Open Source” with free commercial rights.

  • Personal / non-commercial evaluation: OK.

  • Commercial use, SaaS, redistribution, or commercial 二开: paid written authorization required.

  • Full terms: LICENSE.

Available Tools

8 tools
catalogA

列出所有已注册的引擎与当前选择。

Inputs: no named parameters. Returns: result

get_speechB

轮询拉取实时语音事件（speech_start/final/state…）。宿主可据此感知用户说话。

Inputs: timeout, since_id. Returns: result

healthA

返回声络状态摘要（版本、ffplay 是否可用、当前引擎）。

Inputs: no named parameters. Returns: result

listenC

采集本地麦克风语音并识别为中文文本。本地或云端 ASR 均可（云端需先在 config.yaml 配 asr.aliyun）。返回识别结果 JSON（含 text 字段）。

Inputs: engine, timeout. Returns: result

listen_startA

启动常驻语音会话：后台持续听写麦克风并实时转写。返回 session_id 与状态。

Inputs: asr. Returns: result

listen_stopA

停止常驻语音会话，释放麦克风。

Inputs: no named parameters. Returns: result

set_engineA

切换某类引擎（会话级）。kind ∈ vad/asr/agent/tts，name 用 catalog() 查看。

Inputs: kind (required), name (required). Returns: result

speakA

本地朗读文本（可带情绪）。interrupt=True 时打断当前播放，便于实时插话。

Inputs: text (required), voice, emotion, interrupt. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A3.7/5.0

    Scored across 8 tools

    Disambiguation5/5

    Each tool targets a distinct function: speak for TTS, listen for one-shot STT, listen_start/get_speech/listen_stop for streaming sessions, set_engine/catalog for configuration, and health for status. No two tools have overlapping responsibilities.

    Naming Consistency4/5

    Most names follow an imperative verb style with snake_case for compound actions (listen_start, get_speech). A few single-word names like catalog and health deviate from a strict verb_noun pattern, but the overall style is consistent and readable.

    Tool Count5/5

    8 tools is well within the ideal range. Each tool serves a clear purpose in the voice lifecycle (speech, recognition, session control, engine management, health) without redundancy.

    Completeness4/5

    The core voice workflows are covered: one-shot and continuous recognition, TTS, engine switching, and status. Minor gaps exist such as no explicit pause/resume or volume control for speech, but these can be worked around using engine settings.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Resources

    Related MCP Connectors

    Related MCP Servers