Cursor 101: How to get setup for vibe coding
A first-time Cursor user sets up a workspace, creates a planning mode, and runs a small to-do app locally.
Watch on YouTube ↗Follow the walkthrough
1. Install and sign in
Install Cursor and sign into the account used for the walkthrough. Open a project folder so the agent can see and edit the files for this exercise.
Watch at 00:01:002. Choose models by task
Review the model settings together. Amir distinguishes planning and reasoning work from implementation and tool calls.
Watch at 00:04:003. Make a dedicated planning mode
Create the PRD-generator mode shown in the recording and choose a reasoning model. Use it to define the app before switching to implementation.
Watch at 00:13:304. See the result in a browser
Ask the agent to start the local server and inspect the to-do list. Confirm a specific requested detail, such as the yellow styling in the demo.
Watch at 00:20:00
Use one small project to learn the handoff from plan to files to running UI.
Read the transcript
I'm here with Air. Thank you for coming. He's going to change my life today. He's going to graduate me from a VI marketer at level two who's does only automations and gum loop, Zapier, N8, and he is going to level me up to a cursor kid. Listen, by the end of this, I want you to only use cursor moving forward. It's going to be the interface for work, for everything. Let's make this happen. Yeah. When you said that that cursor is going to be the interface for work a
couple weeks ago, I was like I mean it's been stuck in my head ever since and I was like we got to do this. I'm you're going to force feed me cursor turn me into a developer and let's dive in. Let's make this happen. All right. So let's get you to open up your uh cursor window and just get the settings going. Uh I can tell you step by step what you need to toggle. Yeah. So, you're going to want to install git commands and, you know, obviously read the terms
of service. Red. Done. Turn your life away. 9 hours. I don't think it should take 9 hours. That's crazy. Wow. We just spent five hours. Life moves fast in in AI land. There you go. Yeah, we It moves a little differently here. But in the meantime, if you want, while that's installing, um, you should be able to actually log in with your account or sign up if you don't already have. I'm going to go off screen here and make an account. I have
not made an account. Perfect. So, let's go account. I am literally a spring chicken. All right. So, cursor dark is Yeah. Yeah. You want to go with that one. That's a good theme to have. You know what I want? Continue. Oh. Oh, I just saw Greg talking about light mode earlier today. I might have to be savage. I'm a light mode. I'm a light mode machine. Go. I'll let you go with the light mode. No, we got to go dark just just for now. Um I'll learn
about these shortcuts later, right? Yep. I'll show you everything afterwards, what each one means. Shortcuts. Data sharing. Share your data. Let's let's help them out. Wow. All right. And perfect. Yeah, you can ignore that. Okay, cool. Okay. So amazing. You now have cursor fully installed. While we're getting installed, there you go. So this is your interface. Um, in most cases right now with cursor, how it's set up is you need
to have an existing folder that you want to open up cursor into, but I'm sure in the future they're going to let you actually create new folders directly within cursor. Uh, on the right you have your chat window, very similar to what you would see with chat GBT. Um, and you have all these different toggles and I'll explain what each one means. But first, we want to go through settings together and make sure that you are set up for success with cursor. Um, let's
click on cursor on the top left beside the Apple menu and go on settings. And there you go. Cursor settings. Perfect. Okay. So, what we want to do is um right now we have data sharing disabled or privacy mode disabled. and we're going to go to chat and toggle on basically every single everything. Just toggle on everything. The only thing you don't want to toggle is um the last two sections in auto run. So in auto run, yeah, you can click on auto run.
Yep. Continue. Just you want to have everything toggled except the last two. So file protection and external file protection. Yeah, you can untoggle those. Dump these. And then Yeah, exactly. And then toggle everything else. This is vibe toggling. Vibe toggling actually for character level diffs all at the bottom. So the fourth one up you can untoggle. Okay. It's just a personal preference. There's no right or wrong way. These are
just ways that I think could help you move faster. And then you're going to go into the tab section the left underneath chat. Have everything toggled as well. And do I really need to know anything about about what I'm toggling or just No. Take the wheel. Yeah, these are just basically you can toggle everything here as well. These are just made for you to just improve your workflow process. These are just specific let's say micro optimizations or enhancements. Yeah,
you're going to turn that one on as well. All right. And then let's talk about models. So you'll see that there's different models applied here. And my recommendation to you is to think about models in three ways. There's models that are for thinking. They they have very they have reasoning enabled into them and they think very logically and step by step. Then you have aentic models that are more more so for um execution tool calling and actually
doing the work. One or two examples of that. So 03 max only is a great thinking model. It does a great job of being able to actually give you a breakdown of an implementation plan, a requirement documentation or solving very specific hard problems. So if you're facing a specific bug, in most cases I use 03 to come up with a plan as to how we would solve that bug. and then switch to a model that's an agentic model to be able to actually execute on that plan. That
way, the agentic model is not doing the reason you're thinking. It's just following the plan that was provided by O3 model. So, we'll I'll show you what that looks like in action. But for now, what we're going to do is we're going to toggle all the claw models that you have. We're going to un toggle GPT 4.1. And then we're also going to make sure that we have Gemini 2.5 Pro, Opus, Sonnet, uh, and we don't need 3.5 anymore. I think 3.5 is commonly used
but you know with summit 4 now we're in a good place. Amazing. All right. So let's move on to rules and um oh also one other thing with models you can uh you can provide your own um API keys if you you reach your limits or you would like to be able to just have your own uh token token billing and integrated so you're not paying for cursor directly through there. Um but for now you can just have this as is. What do you recommend for
that? I I honestly recommend that you just subscribe to a pro plan for cursor the 20 or $40 plan whatever the price is I think right now and just do direct uh direct billing through them. Uh you don't need to install your API keys only if you hit certain limits. Sounds good. Everyone I feel like is concerned about installing their API keys to save money but I don't think these models are or these groups are making that much money
on tokens. So yeah. So what cursor does if I'm not mistaken is they charge uh so if you if you see API pricing or buildout API pricing um what they do is they just charge a 20% margin on top which is still very like it's it's a rounding error like you won't be able to see the difference. Um I will turn on max mode as well. This is you know maximum context for the models as well and this is priced a little bit differently and we'll talk about when to
use these. Um let's move on to rules. So rules are basically um uh ways or like a set of principles or philosophies that you want to install globally across all your projects. These are maybe tuned to the way of how you would like to work and how you would like the agents to work itself. Um, typically I would say, you know, this was very much um, you know, typically what you saw before was you had rules that just were very holistic level principles like design
clean code, comment your code, make sure it's component based and structured. The way these agents are now, they do that inherently, but it doesn't hurt to have that rule regardless and make sure that it's actually following that. So, uh, in this page, you're going to want to, uh, enable generate memories. It's a it's in beta mode if you're able to do that. Um, oh, this is disabled in privacy mode. So, uh, if you go to back to here, disabled,
uh, I think, yeah, generate memories, it should be, yeah, you should be able to do it. If not, then it's just might be. Yeah. So, go back to where we were in rules. Okay. So, yeah, it's it's fine. It we can come back to that one later on. Okay. Data. Um, all right. Oh, okay. So, it's going to take some time for you to be able to toggle that one on. That's fine. So, keep it disabled. You can keep it disabled if you like. Um, and then in 24
hours, you can come back to rules and turn on that toggle. It's just not available now. I love it. Memories. Got to have it. There you go. And what you're going to do is I'm going to send you in the chat um can I I'm going to text it to you if it's possible. Yeah, because I don't think I can text. Uh, a general rule that I like to have uh applied across all my projects. So you can actually paste that rule in.
That would be amazing. All right, we're going to add rule. Cool. And this just covers the basics around like making sure the code is clean, there's good structure, um, and you're following good core philosophies around code. Sweet. You're going to save it. And sweet. Now, the last thing we're going to do is we're going to go through indexing. And if you click on indexing, you want to have indexing f folders
toggled and docs. So what are docs used for? Say for example, um you're working very with a very specific API uh like exa fire crawl or any other external service. Uh you can actually paste a link to that documentation in your cursor chat what throughout your project and it will automatically crawl and index those uh those documentations for you or you can do it manually at the step right now. So for example, you can just in, you know, put together the
paste the OpenAI docs and it will call for you and and and uh index that as part of a context reference in the chat. For now, we're just going to leave it as is. We can we can cross this bridge later on. And these are account level docs, right? Like I could access across any project. These would like basically my tooling docs. Exactly. Exactly. the end. I'm going to show you a little trick later on as well um where you can just
reference documentation directly and put it as a text file in case if there's any issues with crawling or retrieval. Next up is background agents. This one I'd say is oh you don't have background agents. That's fine. It's just part of the beta plan. But um yeah, if you if you go to um Oh, there you go. Perfect. Enable that one. Okay. Oh, because this the privacy setting that you have, it's gonna take time for it to get enabled. Yeah. So
just we got to toggle it back on. There's too much goodies that I need. Yeah. No. So it has to be actually disabled. But the thing is is that's going to take 24 hours for that mode to get active. Oh, wait. They're not active in privacy mode. Exactly. So you need to have it you need to have privacy mode off. Oh, but the reason why it's not working cuz it takes 24 hours for it to to apply the setting. So I want to stay off. You want to stay off. All right.
Yeah. But you see how on the bottom it says it takes 24 hours. Yeah. Yeah. So, background agents, we'll leave that on for now. But what background agents is, basically, it's fully um asynchronous agentic workflows where you can just one prompt an agent and it will execute the tasks without you having to u do follow up human in the loop prompts to it and it can complete tasks directly. Next up is MCPs. So, MCPS, it's been significantly upgraded. You can now do
one-click installs. They have a directory now. So, if you click on new MCP server, uh you'll be able to actually just install an MCP. We'll talk about what that looks like for now. We're going to leave it. We'll only install MCPS if we need to. Sweet. Okay. I think we've covered the basics. We have um if you go to beta, there's like maybe two other toggles, which is notepads. So, if you you can just install that. Perfect. And early
access. Uh right now it says update access. You have a default. You can do early access. You can get access to new features. Amazing. Okay, sweet. So, you can just close off the settings now. Buy settings. By settings. And you have a folder on your desktop called untiter folder. Let's call that whatever you want. Live jam. Live.
Jam SCP. Perfect. So, you can you can you can now drag that folder into your cursor. You can just move it over and it should open it up. Amazing. Yes. Continue. Close that right there. Amazing. All right. So, we now have a fully working or setup cursor with all the right settings for you to do your best work. We can now get started on doing a bunch of different things. What
would you like to get started on? What what what is your first initial thoughts when you get into it? I have no clue where to start. You have no clue where to start. Um is there so let's say for example um you want to just let's let's do a very very basic to-do list app just to give you an idea of how we would approach something like this. So in your folder you'll see it says live jam MCP. There's about four icons with a plus icon. You got a new file. You're going
to create something called a prd. MD stands for markdown. And a prd is a product requirements document. This is the basis for for me in terms of how I approach builds where I want to first outline exactly how we're going to be building an app or a project. And we talked about how you can use thinking models like 03 to think through a plan of what you would do. In this case, we want to build a very very simple to-do list app.
What we're going to do is we're going to go to the chat on the right side and click on 03. So you have agents. That's right. And then we're going to change the model to 03 the bottom. Now typically what you can do is you can also create custom modes um that are basically specialized agents that are made to perform a specific task. Why don't we go ahead and create a custom mode called uh PRD generator. And we're going to set the model to 03.
And then once you have your purity generator, you can also click on advanced options and it'll let you Yep. And max mode. Um, yep. Purity generator. Yeah, you're going to do 03 and then click uh done or go back. Okay. You're going to advance options and you're going to paste in this prompt that I'm going to provide to you right now. Yeah, you're an expert AI assistant specializing in creating comprehensive product requirement
documents. So now what we're going to do is create our first PRD. So you're going to click done on top right where we have all the settings placed. And we are now selected the PRD generator. In uh in here you're going to type in create a comprehensive PRD. create a comprehensive PRD to build a to-do list app that allows us to make simple task and check them off. And when they're completed, they have a yellow check mark next to them. Perfect.
And all I will say is uh below what you want prompt to say start off with gen uh um outline a very simple MVP using HTML, CSS, and JavaScript code. Boom. And then what you want to do in most cases is sometimes these thinking models don't actually implement the changes. They'll tell you the output but they'll make the change. So what I like to do as a as a safety or backup is just reference that file. So you can actually drag the prd file in the chat even
though it's in context and just yeah say make changes to this document. Boom. All right, perfect. Oh, all right. So, it's fine. So, there's no max. Uh there uh you can switch to let's do let's do turn off the max mode. Yeah. And then just do uh Gemini 2.5 Pro. It won't let you. Yeah. Pro. It's forced max. Gotcha. Yeah. Yeah. Yeah. It's forced maxed. Okay, cool. And now just try again. Oh.
Uh, okay. Yeah, there's no it's forcing the max. So, just do do a model that turn off the max toggle. Yeah. Um, maybe. Yeah, we can do There we go. There we go. Perfect. I definitely recommend for anyone to sign up the pro model for sure. Oh, yeah. Yeah, I'm signing up immediately. No worries. I thought I was on the prot trial, but ran out maybe already. I don't know. I didn't do anything. Oh,
well, maybe it's time based. It's time based trial. Is it? Maybe. I don't know. I got an email that said, "Yo, you're you got a free trial." So Oh, true, true, true. Maybe I just need to activate it. But that could be it, too. cuz like when I was in the settings it said pro trial. Oh, true. Uh I think uh huh whatever. Well, all good. All right. So, so the Gemini model is now thinking through the
steps in terms of what it actually needs to do and it's now going to actually apply that PRD. Uh, and the PRD has a custom mode and prompt that's made to really think about your key users, the user stories, the ICPS, the tasks, and the subtasks. So, what we have is a very detailed breakdown of uh uh an outline of how to actually approach to build an implementation. And you can see it's broken down to key user stories, functional requirements, problem
statements. This is very helpful for I think much more complex applications as well. And you'll see that in sometimes it'll also give you the database schema. So, if you uh scroll down fully, you'll see a full breakdown of kind of how to actually build this out and what the scope will look like. So, you're going to go ahead and click accept. YOLO. There you go. Now, in back in the chat, you're going to change to the normal agent mode,
and you're going to click on instead of Gemini 2.5 Pro, you're going to change it to auto. And this is going to select the best model. Some people just use um Sonet 4, but we're just going to use auto. And then they say uh referencing the PRD generate the simple to-do list app based on the MVP scope. Perfect. So what it's going to do, it's going to reference the PR. It's going to
look at um uh what we specified in the MVP and create a very very simple to-do list app um with just like JavaScript, HTML, CSS. And there you go. We have it. So, and what it's doing is it's not connected to a database. It's not connect connected to anything of that sort. It's completely um all locally stored in your browser. You're going to go ahead and click accept uh in the chat in the right in the chat.
Uh right here. Yeah. Yeah. Cool. So, let's say for example, you weren't happy with this result and you want to go back. So, what you can see in the chat as you scroll up is it's going to have if you scroll up um it's going to have ability for you to restore checkpoint. This is really really helpful if at any point you uh run into an error or you want to redo a prompt or redo your work. You'll see in the previous chat that it lets you actually restore the
checkpoint. So, you can actually go back and click it and start again if you have an issue with it. Now, most of the time people that create this like, okay, cool. the code is built out for me. Now, how do I actually see this? What do I do? Instead of me saying how you can do it, I would love to actually get the agent to help you walk through that process. So, you can ask the agent, all right, how do I actually run this? How do I see
this? All right. Thanks. Thanks for the code. That looks cool, but how do I see it? All right. It's going to set up terminal for you and it's going to run a live server and boom, there you have you have a to-do list. Get MCPS set up. Boom. Perfect. There you go. Nice. Even my little yellow. They got yellow. There we go. So now we have a good understanding of how you can
essentially build something very simple starting from just the approach of coming up with a plan and then getting the agent to execute on it. Now, let's say we want to set up MTPs. Um, in this in we do. Okay. So, let's use my favorite MCP, which is fire crawl. Now, because I'm already set up with some of the API keys, I'm going to give it to you. You can use it. Um, what we're going to do is go to your cursor settings and go to MCPS and yeah, add a
custom MCP. Perfect. Okay. So, cool. So now it's like okay how do we how do we how do we set up MCPS and what this process is. It is a little bit complex but don't be overwhelmed. It's it's very easy to do. Um typically the structure of any MCP um looks uh it's it's just a very basic JSON file that um you would uh call an API key to get your credentials to access the service and then from there you can actually interact with the LM.
I'll take a step back and maybe articulate in a better way. MCPs are what APIs are for regular applications. So if you have a regular app that wants to communicate with an external service to create, update or delete some data, you would use an API. You would have your API keys as your credentials to interface with that service. MCPS are something that was developed by um built by Anthropic, which stands for model context protocol. It's just a new
interface for your LLMs to be able to interface interface with data, capture data, update data, retrieve data, and push data directly through your large language model without actually having to build a custom API call. Do you use Notion? Uh, yeah, sometimes. Okay, let's do notion. Let's do add notion to cursor. Okay, perfect. And install and there you go. Click that. needs
login. Open up. I'll log in. All the game. All right. Amazing. So, you can tell which pages you want to show. Yeah. Select pages. And then you can just choose one of the pages. Perfect. Don't ask me again. Yeah, less toggles the better. Yes. 14 tools. I got tools. There you
go. So, let's open up a new chat. Fresh chat. Let's get started from the scratch. I got fresh tools. All right. Okay. Uh you can now just say using the notion MCP tell me my like last five documents. I would say if we can let's just take out the MCP part for uh fireball completely. Yeah, of course. Yeah. Cool. And then you click on run tool. So, it's going to do a search on the five documents you have.
There you go. You now have direct access to your notion through cursor. And you can use this as a reference to, for example, make changes if you wanted to. So, if you have an existing document um that you want to make changes directly to your cursor and if there's other information that you want to use as well, you can use as a reference point. Sick. So you can view the document blah blah blah. What if I have to like make edits
to it or update data to it? I don't want to because I selected the wrong the wrong page. So is there an artic is there a specific uh document that you want to make edits to? So you can just search it in the chat and say hey using MCP notion MCP find me this document. I want to make some edits to it. Take me to the document. Give me the link. Let's test it. Instead of me going out,
get the link. There you go. Hello from cursor. There you go. You were able to update your notion file directly. Amazing. That's so you built the simple to-do list. You got the MCP set up. You updated notion doc and yeah I think I think got a good uh good foundation. Bro, dude, thank you so much for doing this. This in an hour, less than an hour
really. We started from scratch, signed in to Cursor, we got all the settings dialed. We got a custom mode set up and we built a to-do app. We got an MCP set up for Notion and I mean the the world's our oyster here. You're you're you're you're ready to vibe. I'm ready to vibe. A long way to go, but at least I'm over the hurdle. And thank you so much, man. I appreciate you. It was a lot of fun. Thanks for
thanks for having me. I appreciate it. All right. Where can where can people find you? Amir MXT am I MXT everywhere. Twitter, LinkedIn, that's where you can find me everywhere. Absolute banger videos. So, you gotta you gotta check the videos out and just get like a mirror. Thanks for having me, man. I appreciate it. Peace. Peace.