Kurator MCP Connectors vs Gemini Notebook
Transcript
Experiment Overview: MCP Connectors vs. Gemini Notebook
Speaker: Karan Bavandi
Can Claude and ChatGPT MCP connectors do the same job that NotebookLM or Gemini Notebook does? The Gemini Notebook promise is a simple one: create an isolated knowledge base, ask it questions, create presentations from it, and use it as a foundation for your research. But how would Gemini and Claude MCP connectors connected to that exact same knowledge base perform? Can we get the same value out of them, or is there a noticeable difference between their responses? I just ran through this experiment, and I want to share those results with you.
Building and Structuring the Knowledge Base
Speaker: Karan Bavandi
Before diving into the results, let's take a look at how the knowledge base is created and accessed by Gemini Notebook, Claude, and ChatGPT. The knowledge base I'm using consists of AI Insider weekly webinars where marketers discuss the influence of AI in marketing. Everything is curated directly from YouTube. Each post is then transcribed using custom prompts, nicely formatted with speaker names and titles so that Gemini Notebook, Claude, and ChatGPT can all easily access and make sense of the content.
To connect and sync this information with Gemini Notebook, I use the command directly from Kurator. For Claude and ChatGPT, we have set up MCP connectors so that all of this data is directly accessible simply by mentioning the folder names and executing the prompt.
Cross-Platform AI Comparison: Gemini, Claude, and ChatGPT
Speaker: Karan Bavandi
Now let's look at the experiment results. First, in Gemini Notebook within my AIMG community, I asked: "What are people talking about when they mention 'human in the loop'? Who is human in the loop?" It generated a response structured into clear sections detailing the findings.
I then did the exact same thing in Claude, instructing it to look at the AI Insider folder in Kurator—accessing the exact same knowledge source as Gemini Notebook—and answer the same question. The way Claude formulated and structured its analysis was completely different, and it provided direct links below to all the resources used to generate the answer.
Finally, I asked the exact same prompt in ChatGPT referencing the same folder. ChatGPT's answer was significantly shorter than the others. While it lacked extensive inline links, it provided clean citations pointing directly to the specific videos referenced.
Three different LLMs handled the exact same knowledge base in uniquely distinct ways. As a researcher, marketer, or analyst using AI, each of these varying responses offers a unique insight. It highlights why evaluating across different LLMs is crucial to finding the right solution for your specific task.
Analyzing LLM Analytical Approaches and Differences
Speaker: Karan Bavandi
Next, I wanted to understand why these platforms differ in their outputs. I brought the responses from all three platforms into Gemini and asked it to analyze the analytical differences between them.
The results were revealing. Gemini categorized the platform behaviors as follows:
- Gemini Notebook: High-level conceptual abstraction.
- Claude: High granularity source extraction with clear contextual references.
- ChatGPT: Focused taxonomy mapping with concise summaries.
Each platform uses a distinct processing methodology or master prompt strategy behind the scenes to generate answers. Knowing which platform excels at specific tasks allows researchers and analysts to deploy them far more strategically.
External Operational Memory and Final Takeaways
Speaker: Karan Bavandi
What Kurator offers is the flexibility to use a single content source seamlessly across all of these different platforms. It allows you to create an "operational memory"—a centralized repository outside of any individual proprietary AI platform that you fully own. You can then deploy this memory across various models to perform research, draw conclusions, and generate comprehensive analysis.
Information synthesis and meaning extraction are among the strongest use cases for generative AI today. Leveraging external operational memory across multiple LLMs allows you to get the absolute most value out of your curated data.