podcast-maker

name	podcast-maker
description	Generate an AI podcast (audio + optional video) from any topic using Google NotebookLM. Ingests sources, synthesizes research, generates a natural two-host discussion. Use when user says "make a podcast about", "create an audio summary", "turn this into a podcast", "make me something to listen to".
allowed-tools	["mcp__notebooklm__nlm_create_notebook","mcp__notebooklm__nlm_add_source","mcp__notebooklm__nlm_research","mcp__notebooklm__nlm_generate","mcp__notebooklm__nlm_download","mcp__notebooklm__nlm_list_artifacts","mcp__cf-browser__browser_markdown","WebSearch","WebFetch","Read","Write","Bash"]
model	sonnet
user-invocable	true
when_to_use	When user wants to create audio content, a podcast, or multimedia about a topic. Requires NotebookLM setup (uvx notebooklm login).
argument-hint	<topic or URL(s)> [lang: zh-TW|en|ja] [with-video]

Podcast Maker

You are the podcast creation skill for Claude-Agent. Generate AI-powered podcasts from any topic.

Input

$ARGUMENTS — topic or URL(s), optional language (default: infer from user profile), optional with-video flag.

Prerequisites Check

If notebooklm MCP is unavailable, respond with setup instructions and stop:

Podcast creation requires NotebookLM. One-time setup:
1. Run: uvx notebooklm login
2. Sign in with your Google account in the browser
3. Try again after login succeeds

Steps

1. Parse Input

• Extract topic or list of URLs
• Extract language (default: infer from user profile, fallback to en)
• Check for with-video flag

2. Collect Sources

Gather 2-5 authoritative sources on the topic:

• If URLs provided: use them directly
• If topic keyword: WebSearch for 3-5 relevant articles
• For each source URL: optionally browser_markdown(url) to verify content quality

3. Create Notebook

Call nlm_create_notebook(title="Podcast: [topic]") — returns notebook ID.

4. Add Sources

For each source:

• If URL: nlm_add_source(notebook_id, url=<url>)
• If text content: nlm_add_source(notebook_id, text=<content>)

Add 2-5 sources for a well-rounded discussion.

5. Deep Research (Optional)

Call nlm_research(notebook_id, query="comprehensive overview of [topic]") to enrich the notebook's knowledge base with web sources.

6. Generate Audio

Call nlm_generate(notebook_id, type="audio", lang=<lang>).

This generates a natural-sounding two-host podcast discussion. The generation may take 1-3 minutes.

7. Download Audio

Call nlm_download(notebook_id, type="audio", path="workspace/podcast-[slug].m4a").

Note: The output is M4A format (not MP3), despite what the filename might suggest.

8. Generate Video (if requested)

If with-video flag is set:

1. nlm_generate(notebook_id, type="slides", lang=<lang>) — generate presentation slides IMPORTANT: Generate slides AFTER audio, not in parallel (NotebookLM constraint).
2. nlm_download(notebook_id, type="slides", path="workspace/slides-[slug].pdf") — save PDF
3. Convert to video using Bash:

   pdftoppm -png -r 300 workspace/slides-[slug].pdf workspace/slide-[slug]
   ffmpeg -framerate 1/5 -pattern_type glob -i 'workspace/slide-[slug]-*.png' \
     -i workspace/podcast-[slug].m4a -c:v libx264 -pix_fmt yuv420p -c:a aac \
     -shortest workspace/video-[slug].mp4
   

4. If ffmpeg or pdftoppm are not installed, note this and provide the PDF + M4A separately.

9. Save Notes

Write to workspace/podcast-[slug]-notes.md:

# Podcast: [Topic]
Date: YYYY-MM-DD
Language: [lang]
Sources: [N]

## Files
- Audio: workspace/podcast-[slug].m4a
- Slides: workspace/slides-[slug].pdf (if generated)
- Video: workspace/video-[slug].mp4 (if generated)

## Sources Used
1. [Title](URL)
2. [Title](URL)
...

## Notebook
ID: [notebook_id] (can be used for follow-up questions via nlm_ask)

10. Present to User

Show:

• Audio file path and estimated duration (typically 5-15 minutes)
• Video file path (if generated)
• Source list
• Offer: "Want me to ask the notebook follow-up questions, or create a different artifact (quiz, study guide, slides)?"

Other Artifact Types

NotebookLM can also generate: slides, report, quiz, flashcards, mind-map, data-table, study-guide. If the user asks for any of these instead of a podcast, adapt steps 6-7 accordingly.

⚠️ infographic download is unreliable. Use slides instead for visual content.

Rules

• Always add at least 2 sources for quality discussion
• Generate audio and slides SEQUENTIALLY, not in parallel
• M4A is the actual audio format (not MP3)
• If Bash tools (pdftoppm, ffmpeg) are unavailable for video, provide files separately
• Keep the reply concise — full details go to the workspace notes file