Companion MCP server for VoxAI — it gives your AI ears (live, speaker-labeled meeting transcripts) and a voice (speech on your Mac).
Everything stays on your machine. The server talks to the VoxAI app through its local container; no audio and no transcript ever leaves your Mac through this server.
One command, no sudo — it's a notarized universal binary:
mkdir -p ~/.local/bin && \
curl -fsSL https://github.com/Ethan-YS/voxai-mcp/releases/latest/download/voxai-mcp-macos-universal.zip \
-o /tmp/voxai-mcp.zip && unzip -o /tmp/voxai-mcp.zip -d ~/.local/bin && \
chmod +x ~/.local/bin/voxai-mcpThen register it with your AI client as a stdio MCP server pointing at ~/.local/bin/voxai-mcp.
You normally don't need to do any of this by hand. Open VoxAI → Settings → Connect Your AI → Copy Prompt, paste it to your AI, and it will install and connect itself.
Related MCP server: @vocea.app/mcp-server
The VoxAI app is distributed through the Mac App Store, and sandboxed App Store apps can't ship a helper that an external process (your AI client) is allowed to launch — a bundled executable must carry a sandbox entitlement, and one that inherits it can only be sub-launched by its parent app. Compliant means unlaunchable; launchable means non-compliant. So the server lives here instead, outside the store, signed and notarized under the same Developer ID.
Listening
| check in / out as the user's copilot — the app shows your name |
| poll new segments with a cursor |
| browse and read past sessions |
| give a session a real title |
| surface a note to the user, live, during the session |
Speaking
| talk through the user's Mac |
| what's available (system voices, or Qwen if installed) |
| install the higher-quality voice component on request |
| engine / voice / rate |
Voice chat
| the floating dialog panel: hear the user, reply in a bubble |
macOS 14+
The VoxAI app — this server is useless without it; it reads the app's data and drives its UI
swift build -c releaserelease.sh <version> does the full distribution build: universal binary → Developer ID signing → zip → notarization → Gatekeeper verification.
The server builds and runs on Linux, and it will complete the MCP handshake and report all 14 tools. It cannot do anything useful there: every tool reads or writes IPC files inside the VoxAI macOS app's sandbox container, and that app only exists on macOS. On a non-macOS host each tool call answers with a plain message saying exactly that, rather than silently returning empty data.
docker build -t voxai-mcp .
docker run --rm -i voxai-mcp # speaks MCP over stdio; tools report "needs the macOS app"This exists so that directory sites which verify servers by starting them in a container are checking this server, not a stand-in written to pass the check. If you want the tools to actually work, install VoxAI on macOS 14+ and run the server there.
Proprietary — see LICENSE.
You may download, install, and run this connector on machines you own or control, including at work, when using it together with VoxAI. You may not modify it, redistribute it, or use it to build a competing product.
The source is published here so that anyone can read what a program they run on their own machine actually does. That transparency is the point; it is not an open-source grant.
This server cannot be deployed
Meeting transcripts for AI agents: search calls, read who said what, transcribe files and links.
Screen recording, meeting notes, and voice dictation - all with AI
Meeting bot and transcripts for Google Meet, Teams and Zoom. Live or after, speakers labelled.
Search meetings, export summaries and transcripts, and manage recordings from any AI tool.
