whosaid — Claude Code MCP server: what it does, how to install
Transcribes recordings with speaker diarization locally on Apple Silicon and searches meeting transcripts without sending audio to the cloud.
claude mcp add --env WHOSAID_WORKSPACE=<WHOSAID_WORKSPACE> --transport stdio whosaid -- uvx whosaidWhat whosaid does
The server pairs Whisper speech-to-text with on-device speaker diarization. Instead of returning raw walls of text, it breaks down calls, meetings, or interviews into timed turns attributed to distinct people. An initial voice enrollment allows assigning actual human names instead of generic identifiers, keeping voiceprints in a private registry across future audio runs.
Exposing the tool through MCP gives an assistant access to transcription, speaker relabeling, environment checks, and search across recorded workspace transcripts. Diarization processes audio longer than 15 minutes across parallel slices, keeping processing fast. Everything runs offline: network requests to Anthropic happen solely if Claude is deliberately specified as the transcript summarization engine in the configuration.
The tool functions strictly on Apple Silicon Macs and stops if it cannot locate FFmpeg alongside Python 3.11 or newer.
What it helps with
- separate meeting audio into turns labeled by speaker
- search previous call transcripts for commitments and context
- transcribe recorded audio fully offline on Apple Silicon
- recognize enrolled voices and assign real names to dialogue lines
Example prompts for the agent
- Transcribe
meeting.m4aand list all speakers with their timestamps. - Search the meeting index for any mentions of launch deadlines.
- Relabel
SPEAKER_01in the latest transcript as Alex.
How to install whosaid
Platform CLI
claude mcp add --env WHOSAID_WORKSPACE=<WHOSAID_WORKSPACE> --transport stdio whosaid -- uvx whosaidReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from whosaid <command>
Platform CLI
codex mcp add whosaid --env WHOSAID_WORKSPACE=<WHOSAID_WORKSPACE> -- uvx whosaidReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from whosaid <command>
Config file
{
"mcpServers": {
"whosaid": {
"command": "uvx",
"args": [
"whosaid"
],
"env": {
"WHOSAID_WORKSPACE": "<WHOSAID_WORKSPACE>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from whosaid <command>
Config file
{
"servers": {
"whosaid": {
"type": "stdio",
"command": "uvx",
"args": [
"whosaid"
],
"env": {
"WHOSAID_WORKSPACE": "<WHOSAID_WORKSPACE>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from whosaid <command>
Platform CLI
gemini mcp add -e WHOSAID_WORKSPACE=<WHOSAID_WORKSPACE> whosaid uvx whosaidReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from whosaid <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about whosaid
How do I install whosaid in Claude Code?
Run claude mcp add whosaid -- uvx whosaid or pass a workspace directory using claude mcp add --env WHOSAID_WORKSPACE=<WHOSAID_WORKSPACE> --transport stdio whosaid -- uvx whosaid.
What does whosaid do?
It provides local speech recognition via Whisper, separates speech turns by speaker, maps voices to enrolled names, and searches meeting transcripts offline.
What are the requirements to run whosaid?
It requires an Apple Silicon Mac, ffmpeg, uv, and a Python interpreter at version 3.11 or higher.
