flashscribe-mcp

MCP (Model Context Protocol) server for flashscribe.ai - lets AI agents transcribe audio/video and work with transcripts through the Flashscribe API.

Give agents ears: meeting recordings, voice notes, and videos become text an agent can search, summarize, and act on - backed by Flashscribe's 90+ language transcription pipeline.

Tools

| Tool | Description | |---|---| | transcribe_file | Upload a local audio/video file and transcribe it (waits for completion by default, ~1 min per audio-hour) | | get_transcription | Fetch a job by id: status, detected language, and the full transcript once completed | | export_transcript | Convert a completed job to SRT, VTT, TXT, or CSV - generated locally from cue timestamps, optionally written to a file |

Setup

Requires Node.js 18+ and a Flashscribe API key (fs_live_*, from Settings → API keys - paid plans).

git clone https://github.com/qwertymuzaffar/flashscribe-mcp.git
cd flashscribe-mcp
npm install && npm run build

Claude Code

claude mcp add flashscribe -e FLASHSCRIBE_API_KEY=fs_live_... -- node /path/to/flashscribe-mcp/build/index.js

Claude Desktop (or any MCP client)

{
  "mcpServers": {
    "flashscribe": {
      "command": "node",
      "args": ["/path/to/flashscribe-mcp/build/index.js"],
      "env": { "FLASHSCRIBE_API_KEY": "fs_live_..." }
    }
  }
}

Example

"Transcribe ~/Recordings/standup.m4a and save subtitles next to it"

The agent calls transcribe_file (uploads, waits, returns the transcript), then export_transcript with format: "srt" and an output_path.

Notes

• Files upload as a raw request body (the API rejects multipart). Max 10 hours per file.
• Transcription failures are refunded by the API automatically.
• SRT/VTT/CSV are derived client-side from the job's cues - no extra API calls.

Development

npm test   # builds, then runs unit tests (node --test)

License

MIT