Skip to content

The Curator

AI-powered podcast generation via MCP — give it a topic, get a fully voiced episode back.

Deploy Python 3.12 License: MIT

  • Write the script

    Google Gemini writes a natural, multi-turn dialogue between hosts Annabelle and Link — complete with emotion tags, pacing cues, and genuine banter.

  • Synthesize the audio

    Vertex AI TTS brings the transcript to life with two distinct voices. Raw PCM frames are stitched turn-by-turn into a single .wav file.

  • Store and serve

    Episodes upload directly to Google Cloud Storage. Share the GCS URL or pipe it into your own podcast feed.

  • Secure by default

    Single-user Google OAuth 2.0 via MCP 2.1. Only your Google account can authenticate; session state persists across container restarts.

Quick start

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "the-curator": {
      "command": "npx",
      "args": ["mcp-remote", "https://your-cloud-run-url/sse"]
    }
  }
}

On first use, Claude will open a browser to complete Google OAuth. After that, use the tools directly:

"Create a podcast transcript about the history of radar."

docker build -t the-curator .

docker run --rm -p 8080:8080 \
  -e GOOGLE_CLIENT_ID=your-client-id \
  -e GOOGLE_CLIENT_SECRET=your-client-secret \
  -e ALLOWED_EMAIL=you@example.com \
  -e SERVER_URL=http://localhost:8080 \
  the-curator

pipx install hatch

export GOOGLE_CLIENT_ID=...
export GOOGLE_CLIENT_SECRET=...
export ALLOWED_EMAIL=you@example.com
export SERVER_URL=http://localhost:8000

hatch run start

Where to go next