# Yap MCP [Health: Active]

**Category:** 🎙️ Speech-to-Text  
**Repository:** https://github.com/founden-bot/yap-mcp  
**GitHub Stars:** 0  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/yap-mcp

## Description
Private, on-device speech to text for AI assistants. Transcribe local audio and video files offline with NVIDIA Parakeet TDT v3 (25 European languages, optional timestamps) and search the Yap app dictation history. No upload, no API key, no account.

## Tools
Capabilities this server exposes over MCP:

- **transcribe_audio** — Transcribes a local audio or video file of any length: WAV, MP3, M4A/AAC, MP4, FLAC, OGG, Opus, WebM, AIFF, CAF and more. Detects the language automatically among 25 European languages. Optional start-time stamps per passage and Yap's cleanup (capitalization, punctuation, "um" removed).
- **search_dictations** — Searches the dictation history the Yap app keeps on this computer (Mac and Windows), newest first. Read-only.
- **prepare_speech_model** — Downloads and verifies the speech model ahead of the first transcription, or reports its status.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "yap-mcp": {
    "command": "npx",
    "args": ["-y","github:founden-bot/yap-mcp"],
    "env": {
      "YAP_MCP_CACHE": ""
    }
  }
}
```

**Requires environment variables:** `YAP_MCP_CACHE` — the values above are empty placeholders; fill in real credentials before running (see the repository for what each one is for).

## Documentation

## What Yap MCP does

Yap MCP connects MCP-compatible assistants to two local speech workflows: transcribing recordings and searching dictation history created by the Yap app. The transcription tool accepts local audio or video files, including WAV, MP3, M4A, AAC, MP4, FLAC, OGG, Opus, WebM, AIFF, and CAF. Files can be any length.

Speech recognition uses NVIDIA Parakeet TDT v3 through sherpa-onnx and runs on the computer’s CPU. Automatic language detection covers 25 European languages: English, German, French, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Ukrainian, Swedish, Danish, Finnish, Czech, Slovak, Hungarian, Romanian, Bulgarian, Greek, Croatian, Slovenian, Latvian, Lithuanian, Estonian, and Maltese.

## How it works

The assistant calls `transcribe_audio` with a local file path. The server reads and decodes the audio in memory, then returns the transcript. Optional passage start times can be included, and Yap’s cleanup can add capitalization and punctuation while removing “um”. Audio is discarded after processing; it is not uploaded, logged, or stored.

The first transcription downloads the speech model from Hugging Face and verifies it with a checksum. The download is approximately 670 MB and is the only network request described for normal operation. Later transcriptions run offline. On a modern laptop, processing is reported at roughly ten times faster than real time.

`search_dictations` reads the Yap app’s history database in read-only mode and returns matching entries newest first. Matching text is then available to the assistant, just like other content shared with that assistant.

## Setup and configuration

Node.js 20 or newer is required. `search_dictations` requires Node.js 22.13 or newer. Supported platforms include macOS on Apple Silicon or Intel, Windows x64, and Linux x64 or arm64.

For a standard MCP installation, configure a stdio server with command `npx` and arguments `-y github:founden-bot/yap-mcp`. Claude Desktop also supports the downloadable `yap.mcpb` package. Claude Code can add the server with `claude mcp add`, and Gemini CLI can install the repository as an extension. A Claude Code plugin is available and includes a `transcribe-recordings` skill.

The `YAP_MCP_CACHE` variable changes where the model is stored. Defaults are `~/Library/Caches/yap-mcp` on macOS, `%LOCALAPPDATA%\\yap-mcp` on Windows, and `~/.cache/yap-mcp` on Linux. On Windows, an existing model downloaded by the Yap app is reused.

## Tools and capabilities

- `transcribe_audio` processes local audio or video, detects the spoken language, and optionally returns timestamps and cleaned text.
- `search_dictations` searches Yap dictation history without modifying it.
- `prepare_speech_model` downloads and verifies the model in advance or reports its current status.
- The same transcription engine can also be run from a terminal with `yap-transcribe`, including timestamp output and multiple input files.

## Limitations and notes

Processing is local, but the initial model download requires network access. The model is downloaded once before offline transcription is available. Search history is tied to the Yap app database on the computer and is read-only. The speech model is NVIDIA Parakeet TDT 0.6B v3 under CC-BY-4.0; its ONNX conversion is from sherpa-onnx under Apache-2.0, while the server code is MIT-licensed.

_Full upstream README: https://allmcps.com/mcp/yap-mcp/readme_