newtype.
ArticleOctober 11, 2025Free to read

Cursor Is the Strongest Knowledge Base App, Bar None

As a programming tool, Cursor is born with RAG capabilities and can index and process local files, making it a powerful knowledge base app.

Originally published . English translation: . Read the Chinese original.

Original video in Chinese.

Key Takeaway

  • As a programming tool, Cursor is born with RAG capabilities and can index and process local files, making it a powerful knowledge base app.
  • Cursor pairs perfectly with Obsidian’s local storage nature, allowing users to directly operate on and make use of Obsidian’s note files through Cursor.
  • By creating a cursorrules document, users can customize how Cursor works, such as prioritizing retrieval of local documents, performing online searches, and calling specific MCPs (such as Markitdown for PDF conversion and Sequential Thinking for breaking down complex problems).
  • Cursor’s Composer and chat features can summarize, distill, and expand notes, acting as Obsidian’s “super AI plugin.”
  • This combined workflow can balance local knowledge and online information, provide logically strong answers, and greatly improve knowledge management and creative efficiency.

Full Content

The best knowledge base app is Cursor, bar none!

If you often watch the videos I post, you’ll know that over the past year or so, I’ve used way too many knowledge tools. The more I used them, the more I realized: simplicity is the ultimate sophistication. Just use Cursor.

Think about it: Cursor already has RAG capabilities. It’s a programming tool. Since it has to program, it definitely needs to understand the state of all the code. So when Cursor opens a folder, it indexes and hashes all the documents inside it. For files like Markdown, it chunks them and embeds vectors.

So the RAG capability that a knowledge base app should have is something Cursor is born with. Whether it’s code or plain text, it handles them the same way. And because it has to write code, it must also have the ability to create and modify documents. That means it can help us write notes and articles directly, right?

That’s not all. Cursor itself has search capabilities. You don’t need to add any special tools; it can search the web and even open a webpage directly.

Finally, whenever the most advanced models come out, Cursor always supports them first. So this $20 subscription, in my personal opinion, is extremely, extremely worth it. Of course, if you don’t want to use Cursor’s model, you can also put in your own API key.

So how do we use Cursor to work with our own documents?

My experience is that you absolutely must create a cursorrules document. Put this document in the root directory and use it to tell Cursor what project rules it must follow. Let me show you.

Look, in the document I specified:

First, before answering, it must always retrieve all documents in the folder first and see whether there’s any relevant content that can be used as context.

Second, checking local documents alone isn’t enough; it also has to search the web so the information is complete.

In addition to these two basic requirements, I also gave Cursor two MCPs:

If it runs into a PDF document, use the Markitdown MCP to convert it. Otherwise, Cursor would have to write a Python script to handle it, which would be very troublesome.

If the problem is a bit complex, use Sequential Thinking to break it down and organize the content, so the logic is stronger.

I wrote out the use cases and trigger conditions for these two MCPs very clearly for Cursor.

Finally, I also attached an example showing how to handle each step, clearly and explicitly. AI can absolutely understand it.

With this whole set of underlying rules, Cursor knows how to work with me.

For example, I ask it: how do you locally deploy a knowledge base?

First, it spends a few seconds thinking about the problem, because I turned on the Thinking option.

Then, it retrieves the documents in the folder.

Then, it searches the web to further supplement the information.

Finally, it uses Sequential Thinking to sort through all the content and organize the logic clearly.

You see, using this method, after this kind of workflow runs, both the content in the knowledge base and the content on the web are taken into account, and the answers you get are also highly logical. That’s why I say Cursor is the strongest knowledge base app.

The rules I just demonstrated are my own needs. Everyone can modify them according to your own needs. In fact, you can even have Cursor help you write them.

OK, that’s it for this episode. If you want to discuss AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!