Whisper Mcp
Give Claude or Cursor on-device speech-to-text so meeting clips, voice memos, and podcast drafts become searchable text without sending audio to a cloud API.
Overview
io.github.jwulff/whisper-mcp is an MCP server for the Build phase that transcribes audio locally using whisper.cpp and Whisper models for agent-driven workflows.
What is this MCP server?
- Runs transcription locally via whisper.cpp—no OpenAI API required for inference
- stdio MCP transport (npm package whisper-mcp, version 0.1.1)
- Uses OpenAI Whisper model weights with local execution
- Suited to privacy-sensitive interviews, internal standups, and offline environments
- Pairs with content and support workflows that need verbatim quotes from audio
- MCP server version 0.1.1
- 1 npm stdio package: whisper-mcp
- Local inference via whisper.cpp (no cloud STT requirement stated in registry)
Community signal: 2 GitHub stars.
What problem does it solve?
You have hours of audio and need accurate text in your repo or tickets, but cloud transcription costs, latency, and privacy rules make pasting files into chat untenable.
Who is it for?
Indie builders who process interviews, Looms, or voice notes inside Claude Code and want offline-capable Whisper transcription.
Skip if: Teams that need real-time streaming captions, speaker diarization as a managed service, or zero local setup—use a hosted STT API instead.
What do I get? / Deliverables
Your agent can transcribe local audio files on demand so transcripts land in plans, docs, and prompts without a separate SaaS tab.
- Plain-text transcripts from local audio files returned through MCP tool calls
- Agent-readable dialogue suitable for summaries, tickets, and documentation drafts
Recommended MCP Servers
Journey fit
Transcription is an integration layer you wire into agent workflows while building products, docs, and content pipelines—not a launch or growth tactic by itself. stdio MCP plus whisper.cpp is classic agent-tooling: your assistant calls local Whisper models instead of you manually exporting subtitles.
How it compares
Local whisper.cpp MCP integration, not a hosted OpenAI Whisper API skill or a video editor.
Common Questions / FAQ
Who is io.github.jwulff/whisper-mcp for?
Solo builders and small teams using MCP-enabled agents who need private, on-machine transcription from meeting and content audio.
When should I use io.github.jwulff/whisper-mcp?
Use it during build and content workflows when you want the agent to read transcripts from local files without calling a cloud speech API.
How do I add io.github.jwulff/whisper-mcp to my agent?
Register the npm package whisper-mcp (0.1.1) as a stdio MCP server in Claude Desktop, Claude Code, or Cursor, ensure whisper.cpp dependencies and model files are installed, then invoke transcription tools from the agent.