Free and open source · MCP server
Speech to text for your AI assistant, on your computer.
Yap MCP lets Claude, Cursor, VS Code and any MCP client transcribe your recordings and search what you dictated with Yap. Recognition runs locally with the same speech engine as the Yap app: no upload, no API key, no account.
npx -y github:founden-bot/yap-mcpTools
What your assistant can do with it.
transcribe_audio
Transcribes a local audio or video file of any length, with optional start-time stamps and Yap's text cleanup.
search_dictations
Searches the dictation history the Yap app keeps on this computer, newest first. Read-only.
prepare_speech_model
Downloads and verifies the speech model ahead of the first transcription, or reports its status.
Ask things like “transcribe ~/Downloads/interview.m4a with timestamps”, “summarize the voice memo on my desktop” or “what did I dictate about the budget this week?”. It reads WAV, MP3, M4A/AAC, MP4, FLAC, OGG, Opus, WebM, AIFF, and CAF and more, of any length.
Install
Add it to your client.
Needs Node.js 20 or newer (22.13+ to search dictations). Runs on macOS (Apple Silicon and Intel), Windows x64, and Linux x64 and arm64.
Claude Desktop
Download yap.mcpb and open it (macOS and Windows), or add this to claude_desktop_config.json:
{
"mcpServers": {
"yap": {
"command": "npx",
"args": [
"-y",
"github:founden-bot/yap-mcp"
]
}
}
}Claude Code
claude mcp add yap -- npx -y github:founden-bot/yap-mcpCursor, VS Code, Windsurf, Cline, Zed and others
Add a stdio server with the command npx and the arguments -y github:founden-bot/yap-mcp.
The first transcription downloads the speech model once (about 670 MB) and answers with its progress meanwhile. After that it works offline. In our test on a MacBook Pro with Apple Silicon it transcribed 41 seconds of speech in 4 seconds, about 10 times faster than real time.
Privacy
Your recordings stay put.
- Audio is decoded in memory and discarded. Nothing is uploaded, logged or stored.
- search_dictations opens the Yap history read-only and returns only what matches.
- The code is public on GitHub under the MIT license, so you can check every line.
Questions people ask.
- What is Yap MCP?
- A free, open-source MCP server from Yap. It lets an AI assistant transcribe audio and video files on your computer and search what you dictated with the Yap app, without sending audio anywhere.
- Does my audio leave my computer?
- No. The file is read from the path you give, decoded in memory, transcribed with NVIDIA Parakeet on your CPU and discarded. The only download is the speech model itself, once (about 670 MB, checksum verified). The text then goes to the assistant you asked, like anything else you show it.
- Do I need an API key or an account?
- No. Yap MCP needs Node.js 20 or newer and nothing else. It is MIT licensed; the speech model is CC-BY-4.0.
- Which languages does it understand?
- 25 European languages, detected automatically: English, German, French, Spanish, Italian, Portuguese, Russian, Ukrainian, Croatian, Slovenian, Latvian, Lithuanian, Estonian, Finnish, Swedish, Danish, Dutch, Polish, Czech, Slovak, Hungarian, Romanian, Bulgarian, Greek, and Maltese.
- Do I need the Yap app?
- Not for transcription. The search_dictations tool reads the history of the Yap app, so it only finds something once you dictate with Yap on that computer.