notebooklm-query

Access a NotebookLM notebook, ask questions via chat, and retrieve information.

Input

The user provides:

• •NotebookLM URL (e.g., https://notebooklm.google.com/notebook/xxxx)
• •Question or research topic (one or more)

If arguments are missing, use AskUserQuestion to clarify.

Instructions

Step 1: Get Browser Context

1. •Use tabs_context_mcp to get tab info
2. •If the specified URL is already open in a tab, use that tab
3. •Otherwise, use tabs_create_mcp to create a new tab and navigate to the URL
4. •Wait 5-10 seconds for the page to fully load

Step 2: Verify Page Readiness

1. •Take a screenshot to check the page state
2. •Confirm the NotebookLM chat input field is visible
3. •If login is required, notify the user and wait

Step 3: Send Question

Craft an effective question from the user's research topic and send it to NotebookLM.

1. •Use find to locate the chat input field
2. •Use form_input to type the question text
3. •Use find to locate the "Send" button and click the one near the chat input
   • •Note: There may be multiple send buttons (source search and chat). Use the one closest to the chat input.
4. •Wait 15-25 seconds (response generation takes time)

Step 4: Retrieve Response

1. •Take a screenshot to confirm the response is displayed
2. •Use javascript_tool to extract the response text:
   javascript

   const chatArea = document.querySelector('[role="log"], [class*="chat"], [class*="conversation"]');
   if (chatArea) {
     const text = chatArea.innerText;
     const lastQ = text.lastIndexOf('<partial question text>');
     if (lastQ >= 0) text.substring(lastQ, lastQ + 5000);
   }
   

3. •For long responses, adjust offset and retrieve in multiple passes
4. •If an expand button (▼) appears at the end of the response, click it before re-extracting

Step 5: Follow-up Questions (if needed)

• •For broad research topics, send follow-up questions based on initial responses
• •Repeat Steps 3-4
• •Maximum 5 questions per skill execution

Step 6: Organize Results

Present the retrieved information in this format:

## NotebookLM Research Results

### Source
- Notebook: [Title](URL)
- Sources: N items

### Q1: <question>
<response summary>

### Q2: <question>
<response summary>

### Key Takeaways
- Point 1
- Point 2
- ...

Tips

• •NotebookLM responses can be long; use javascript_tool with varying offsets for multiple retrieval passes
• •Extract code examples and JSON accurately
• •If NotebookLM is still generating (loading indicator), add extra wait time
• •If chat history is long, the input field may scroll off-screen; scroll down before interacting

Error Handling

• •Tab closed: Re-fetch with tabs_context_mcp and re-access in a new tab
• •Login required: Ask the user to log in manually, then resume
• •No response generated: Wait an additional 15 seconds; if still empty, retry submission (max 2 retries)
• •Send button not found: Use read_page to list interactive elements and identify the button