I've seen so many people struggling with .cursorrules and with the new .cursor/rules directory.
Here is a Guide for you! 🧵
1/ Understand Cursorrules
A lot of my teammates, including myself, always forget to do certain things before leaving the office.
So, we set up our own rules and routines to make sure we do these things every time we leave.
This is basically how .cursorrules work.
2/ .cursorrules file
A rules file is like a guidebook for your AI coding helper. It tells the AI how to write code for your project, including what tools you're using and how everything is organized. This helps the AI create better and more accurate code.
- Create a file called ".cursorrules" in project root.
- Add your rules
3/ Now, let’s go back to our real-world office example. What if we had rules for the kitchen too? In that case, we wouldn’t write them on the same note as the office rules. Instead, we’d put them on a separate piece of paper and stick it to the kitchen door.
So, each space has its own rules.
4/ Before, we had to put all rules in a single .cursorrules file. Typescript, Database rules, UI everything in one place. But that wasn’t ideal. Sometimes, the Agent didn’t know which rules to apply, and you couldn’t be specific. Technically, it just filled up your context window with unnecessary information.
5/ This is how you add your "main" .cursorrules file.
Cursor reads this file first when the AI is working – it’s the first piece of context it loads.
8/ Now you need to reimagine your .cursorrules files too.
They still work like before, but when using them with an agentic approach, it’s different. Now, you're telling the agent how to act, not just listing rules.
9/ A simple example .cursorrules file
(Just that you get the idea)
10/ You can do to file referencing with the @ symbol
11/ Agent Mode
The best thing is that you can build your full autonomous agent with this approach. For this you need to enable the Agent Mode in Cursor and describe in your rule files what script or what document should be processed.
12/ Want to build and ship faster with Cursor AI?
Get instant access to my Ultimate Cursor Course – Early bird discount available (only in february)! 🚀
New update coming soon: Agent Mode, new cursorrules and example lessons!
AI Prompts (Cursor Rules/Prompts) has already its small ecosystem:
✅ Cursor Extension live (Free)
✅ Web Directory incl. Search / Filter (Copy Paste)
✅ Open Source on Github
🔁 Create your own Rules WIP
🔁 MCP Server WIP
... more
MCP (Model Context Protocol) is getting tons of attention on the tech side.
To really understand its impact, we need to see how Tool Calling Agents can make a big difference for companies and especially their non-tech employees in the future.
Thread 🧵
1/ Scattered tool use
Most AI interactions today happen separately. You ask one AI tool for one specific thing. And honestly, even this isn't common yet. Usually, people still have to open multiple tabs or tools, paste data back and forth, and manually search through apps like Slack, HubSpot, or GitHub.
And that's messy. Data and answers get scattered across different platforms. This leads to confusion and bigger expenses.
So, right now, working with multiple isolated AI tools is slow, expensive, and frustrating.
2/ Natural language and one chat app.
That's why ChatGPT or Claude feel easy. Multiple Saas Tools (CRM, ATS, CMS, name it!) or multiple AI assistants add a lot of complexity.
The goal should be a single chat that calls them silently. This isn't MCP's main job, although it can lead in this direction as it handles all these tool connections in a smooth way. (See Claude Desktop)