TranscribeMCP

MCP server for GovTech's Transcribe speech-to-text service. Lets an MCP client (Claude Code, Claude Desktop, etc.) upload audio, run batch transcriptions, and work with the results — summaries, minutes, sections, notes, and transcript Q&A chats.

Setup

1. Get a Transcribe API key: log into https://www.transcribe.gov.sg/ and visit https://www.transcribe.gov.sg/dev_api_key (or use the POST /auth/otps + POST /auth/tokens + POST /auth/apikeys flow if your org isn't on WOG-AD). API keys are valid for 90 days.
2. Install the package:

   pip install -e .
   

3. Set the required environment variables:
   • TRANSCRIBE_API_KEY — your 90-day API key
   • TRANSCRIBE_EMAIL — the email address the API key was issued to
   • TRANSCRIBE_BASE_URL — optional, defaults to https://core.transcribe.gov.sg
   • TRANSCRIBE_API_VERSION — optional, defaults to 3.0

Registering with Claude Code

claude mcp add transcribe \
  --env TRANSCRIBE_API_KEY=your-api-key \
  --env [email protected] \
  -- transcribemcp

Or add directly to .mcp.json:

{
  "mcpServers": {
    "transcribe": {
      "command": "transcribemcp",
      "env": {
        "TRANSCRIBE_API_KEY": "your-api-key",
        "TRANSCRIBE_EMAIL": "[email protected]"
      }
    }
  }
}

Registering with Claude Desktop

Add to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "transcribe": {
      "command": "transcribemcp",
      "env": {
        "TRANSCRIBE_API_KEY": "your-api-key",
        "TRANSCRIBE_EMAIL": "[email protected]"
      }
    }
  }
}

Scope

Covers batch transcription (workspaces, projects, transcriptions), summaries, minutes, sections, notes, and chat-based Q&A over a transcript. Live Transcription is not supported — it requires real-time WebRTC audio streaming, which doesn't fit an MCP tool-call model. See docs/superpowers/specs/2026-07-05-transcribe-mcp-server-design.md for the full design rationale.

Development

pip install -e ".[dev]"
pytest