Podcast Summarizer MCP 📝

MCP that works with OpenClaw, Claude Desktop, Claude Code. Search & discover YouTube channels via natural language and summarize videos with no transcripts or subtitles required.

Quick start (5 minutes)

Python 3.10+ and a Gemini API key (free) from https://aistudio.google.com.

1. Install

pip install podcast-summarizer-mcp

This puts podcast-summarizer-mcp on your PATH. (Or use pipx install / uvx install if you prefer isolated tools.)

2. Configure ONE host

Pick whichever you use. Replace AIza... with your Gemini key.

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "podcast-summarizer": {
      "command": "podcast-summarizer-mcp",
      "env": { "GEMINI_API_KEY": "AIza..." }
    }
  }
}

Quit Claude Desktop fully (⌘Q) and reopen. Click the 🔌 icon — podcast-summarizer should be listed.

Claude Code

claude mcp add podcast-summarizer \
  --env GEMINI_API_KEY=AIza... \
  -- podcast-summarizer-mcp

claude mcp list   # should show podcast-summarizer

OpenClaw

Add to ~/.openclaw/openclaw.json under mcp.servers:

"podcast-summarizer": {
  "command": "podcast-summarizer-mcp",
  "env": { "GEMINI_API_KEY": "AIza..." }
}

Restart OpenClaw (pkill -f openclaw-gateway; openclaw).

Installing from source instead? git clone, cd, then python3 -m venv .venv && .venv/bin/pip install -e ., and use the absolute path $(pwd)/.venv/bin/podcast-summarizer-mcp for the command field above.

Related MCP server: YouTube Insights MCP Server

Example prompts

Discover:

Does Andrej Karpathy have a YouTube channel?

Find me a few investing podcasts.

Track:

Add Forward Guidance to my channels

Add @ForwardGuidanceBW

What channels am I tracking?

Remove Forward Guidance

Summarize:

Summarize this video: https://www.youtube.com/watch?v=MO9ZTZPUwXY

Summarize today's new videos from all my channels in parallel

I have a 20-video backlog — no rush, do it overnight to save cost

The agent picks analyze_video_start (parallel, full price) by default and analyze_videos_batch_start (50% off, async) only when you say "no rush" / "overnight".

📱 Use via Telegram, WhatsApp & More (OpenClaw)

Connect this MCP to Telegram, WhatsApp, Discord and 20+ messaging platforms via OpenClaw — a self-hosted AI gateway. Talk to your podcast research agent from your phone, anywhere.

Telegram → OpenClaw agent (Claude / Gemini / GPT) → podcast-summarizer-mcp → Gemini + YouTube

OpenClaw routes messages from your chat platform of choice to an AI agent. The agent talks to this MCP over standard stdio — no Python wrapper or shim required.

Setup

# 1. Install OpenClaw (Node 22+)
npm install -g openclaw

# 2. Add a Telegram bot token (interactive — paste BotFather token)
openclaw configure --section channels

# 3. Add this MCP + agent + model to ~/.openclaw/openclaw.json:
#    (already covered in Quick Start — use the OpenClaw snippet)
openclaw config set agents.defaults.model "anthropic/claude-sonnet-4-5"

# 4. Start the gateway
openclaw gateway

Tools

13 tools. The agent picks; you don't call them directly.

Registry state lives at ~/.podcast-summarizer-mcp/channels.json.

Configuration

Set in the host's env block. Only GEMINI_API_KEY is required.

VIDEO_ANALYSIS_PROMPT_PATH is re-read on every analysis call (no restart). Bundled examples in prompts/: investment-podcast.md, technical-talk.md, interview.md, news-briefing.md.

MIT License.