Transcription
Hello everyone. My name is Kostya, and this video is dedicated to Codex. Here we will talk about how to use it, how to install it, what types of Codex there are from OpenAI, how to install MCP Skills, how to create a specification, so-called technical requirements for a neural network, according to which it will create your project, and other additional features.
Currently, OpenAI provides access to its ChatGPT and Codex absolutely for free. I don't know how long this will last. It says a limited amount of time, but at the moment, using a simply created free account, you can use Codex and develop using, for example, GPT 5.3 Codex. Also, version 5.4 of GPT has already been released. This means you can use it completely for free.
Let's talk about Codex itself. It looks like this. It is available immediately on both Windows and MacOS. So, here is the interface where you type into the chat what you want to get. It, accordingly, writes code, and you can even see it. You can also run several projects simultaneously. Perform any calculations within them. Skills are available, so-called MCP servers are available, but we will talk about them a little later. It explains here what is available in more detail, I will leave a link, you can read it.
Also, in addition to this desktop version, there is a CI program. It is installed using the following program, i.e., in your terminal. To run the program in the terminal, you must download Node.js. It will not work without it because the npm manager is used here. So you will need to copy this command, open your terminal. Then you will need to paste it here like this and press Enter. Since I have MacOS, I will need to type sudo here. You most likely won't need it on Windows, just npm and so on. But before that, make sure to install the very latest version.
Next, to launch it, just type Codex. Here, you must, or rather, log in. You will need to enter your login here to authenticate. I think it will ask you to do this immediately. I have the following models available. I can show you which ones, so these models are available to us now on the free plan. So I have a free plan, not a paid one, but in this video, we will not discuss the strengths. It is, so to speak, not the most convenient to use, in fact. It's better to use Codex, it looks like this, it's much more convenient. Or, alternatively, you can use an extension for Visual Studio Code. This is a special program for code editing, where you can install the Codex extension. This is done as follows. Type Codex here, it will find it among the extensions, and you will click here and install it. I already have it installed, so I can only disable or delete it. Everything here will be roughly the same as in the regular Codex program. So you can also select models, their level of thinking. There are also logout settings. You can log into your account. I have already logged into mine. I have certain limits available. Also, if the free account is not enough for you, you can go to the official website gpt.com in the pricing section and buy it using any foreign card. The tariffs here are quite cheap. It will cost approximately 100 rubles, plus it will be enough for you for a normal amount of coding. This is not the same story as with Cloud Code. In Cloud Code, on the basic tariff, you can do almost nothing. It's 30 minutes of coding and that's it. Here, it's enough for a much longer time, an hour or two, or even more, depending on which model you use and what level of reasoning. But if you don't have a foreign card, there is this service Plat Market. They don't pay me anything for advertising. This is their big oversight. It's time to start paying, guys. Contact me, perhaps. So, accounts are sold here, buy them at your own risk, but they are quite cheap and generally usable. I think other viewers of my channel might write to you in the comments about these accounts, because I know that many people use them and everything is okay with them.
So, let's move on to Codex. I think you can log in here without me. It's all done very easily. Just log into your account, authenticate through Google, and that's it. Let's move on to Codex. After you install it, it will immediately prompt you to log in. You can do this through your ChatGPT account or through an API. I don't recommend using an API. Just log in through your account. So you will need to continue through Google or through Apple, through Microsoft. I think it's clear here. I will continue through Google, and that's it, my authentication is complete. I can use my account.
Next, look at the limits. You can click on settings. By the way, you can change the application language to Russian if you are interested. Also to Kazakh, Ukrainian, Belarusian. I think they are all here. So, for example, to Russian. Let's see our limits. It says we have 52% of the weekly limit remaining, which will be updated on March 15th. Accordingly, in about 6 days, it will be updated again. I will be able to use all my limits again. But I don't know about others. I switch accounts, well, between them. I have enough of them, I don't get any bans. They say that some people still get banned if they do this. I don't. Maybe I'm special, I don't know, but I'm not banned anyway. But I don't use more than five accounts. I use a maximum of two.
To add any of your projects here, you need to go to your computer and drag some folder here. For example, I will drag Settings, Codex settings, and inside you can start a new dialogue. Accordingly, Codex can also work with Git repositories, if you know what that is, i.e., where all your code is stored. If you don't have GitHub, be sure to register it. Also, if you are interested in me recording a video about GitHub, how to properly store code, how to make conditional commits there, updates. That is, for example, you have developed some feature and you need to add it there. I can show you all of this, how to use all of this, how to properly store code, and so on.
Regarding controls. We have a standard input here, where we can enter any words. We have voice input, which understands Russian, which is very good. Here we can select the model that interests us. No, for example, 53. So, I will use 53 Codex. Here we can select the level of reasoning. What does the level of reasoning affect? It affects the elaboration of your task. And not always very high, it's not necessary to set it. In some cases, it can even be harmful. Here we can add planning. That is, before creating an application, it can first create a plan for it, or work out some task. That is, first it plans it, then it does it. This is very convenient. We can also select the operating speed here, it's better to choose standard. If it's a small task, a correction, you can choose fast.
Next, we also have MCP skills here. I will talk about them a little later. We can also add some files, photos that will help you in development. Next, there are access levels here. By default, when you confirm what the neural network does, there is full access. When it understands itself what to do in what sequence, which, for example, terminal commands to run, it is somewhat dangerous. It is written here that there is an increased risk because it can do something that it misinterprets your command, for example, delete your entire disk. Such cases have happened, but not with me, for example. I usually set full access and, well, I write the prompt in such a way that it is not ambiguous, so that it doesn't go to another disk and delete some software on my device. We also have a terminal and a panel here where you can see what changes it is making to the project. Like this. Here you can select modes and so on. You must connect Git repositories. In the console, you sometimes need to type something, for example, to run a project. For this, a console is needed.
Let's talk about MCP now. In the settings, we can go to settings. And here we have MCP servers. We have pre-installed ones like Playwright, for example. So it can automatically use your browser for testing and development. Just as it is written here. So you can just tick the box, install. Well, these are pre-installed, they are all ours, we will add them too. So, at the moment, Playwright will work.
Regarding MCP in general, we have this website called MCP Market. We have the top 100 MCP servers here that can be used. So we also have our Playwright here, which is already connected. There are others too. You can read about each one yourself, what it's for. There is a description here, but in general, these are so-called enhancers. For example, Siren is a must-have in any of my projects. I use it in every project. Look, so you have a neural network. With the help of MCP servers, it can do some things much better, or get access to something. For example, we can connect Supabase to an MCP that will understand what is happening in your database. It can create new tables, create certain rules for tables, help you create authorization, and so on. So it will understand, for example, context seven is access to the entire modern library of so-called documentation for programming languages, for various libraries. So the most up-to-date information on documentation will be stored on the MCP server, and your neural network can send requests and receive them. And thus, the neural network will always have the latest documentation and perform any tasks much better. Often, well, look, for example, on this site, there is a link to GitHub, or rather, to npm. Let's go to GitHub. Look, there is all the instructions on how to install all this. I will show this using the example of ChatCDN. This is a special MCP server that stores various components for development and so on. Themes. In general, it's a visual enhancer for your projects. We will install it today. We have the Quickstart section here. We have this command. We need to copy it. Then let's go back to our Codex. Look, for example, Playwright, we can look at the settings, how it is installed. Look, it says NPX, then, Playwright, MCP, and so on. In the arguments, it says the command to run. We will have everything the same here. So, when we add our server, it says the command to run. We will also type, or rather, even copy NPX from here. And as an argument, we will put this in and write here, for example, that this is ChatCDN. And that's it, we save it. So now we have ChatCDN. We can use it in our projects, we can turn it on, we can turn it off. So the settings are like this. I think it's generally very easy to install here. So in other, for example, any open codes, it's much more complicated. And in Codex, it's super simple.
Also, besides MCP, we have skills, so-called skills. What are skills? These are conditional prompts for working with something, i.e., pre-installed. We can choose from ready-made ones here, for example, some Playwright skill. We also have a specialized website where we store all sorts of skills that we might need in our work. For example, here we have a skill, a specific one. Next Cash Component Expert. You can translate the page into Russian and read what it's for. For example, this skill provides specialized recommendations for implementing NextJS caching components, i.e., it helps you in your work. How does this skill look? Essentially, it's one huge sheet of recommendations, text, on how the neural network should do things, with explanations. Essentially, it's just a prompt, just a large one, so that it can use something to cache, as it is written there. And there are many such MCP servers. We can simply, for example, download this skill. It will appear in our downloads, and we can install it in Codex. To install it, we type new skill. Here, Skill Creator appears. And then we just drag the entire folder with these skills. So here are references, patterns, skills, the skill itself. And we just tell it: "Add, add a new skill from the folder." And thus, it will now add this skill to us using Skill Creator. This method works, but it's quite slow, and you spend tokens. There is a much simpler method. So you need to copy your skill component in my case, well, your skill where this downloaded from the site is stored. Then, in my case, since I have MacOS, I need to go to my main folder, find Codex here, here we have skills, and just paste it here. Then I go to skills. It will already be here. Accordingly, I can turn it on, turn it off, and configure it. Like this.
Next, in Windows, it will be a little different. You have, you have hidden folders in your user documents. You will need to view hidden folders. You should have the exact same .codex folder, in which you will store everything the same. If you don't have such a folder, you can, in principle, ask your Codex where this folder is located, so that it writes you the path. So write it here in the chat, ask it.
Now I want to show you how to create a specification. Technical requirements, call it what you want. For creating projects. I have such a file called Promgen. I drag it into ChatGPT. It's better to use the web version, not Codex. It will handle this task better here. Ideally, use Cloud for this. So, by the way, Promgen, this file, I will leave in Telegram. You will find it in the first comment. So, I ask, help, help create a specification, a specification, for a NextJS project. And I don't know, for example, a pizzeria, a single-page one. Then it will read this Promgen MD, and it will know my name. And now it will clarify with me. So, it immediately, in principle, started creating this specification based on Promgen, because it understood the project immediately. Usually, it might ask you some questions, three or four. So, now it has started creating the specification, so now it will write it completely. Create a file for me. This is this specification. Now it will simply take this entire specification and package it into an MD file. You can also ask it, for example, well, it will figure out on its own what needs to be written in English. In fact, the fact that it does it in fragments like this is not a problem. Now it is creating all this. It's writing footer, SEO, elements. Apparently, it has finished, I understand. Yes. Now we can download this MD file. Uh-huh. We can't. Okay, that's not a problem. In this case, we will just open Visual Studio Code, drag it here, drag the Codex settings folder here, for example, and create a new file here, for example. And write here project.md and take this whole spec. Copy it like this, the whole response. Drag it here and remove this. Like Kostya. Blah blah blah. We don't need this. So, why didn't it copy all of this? We copy all of this and paste it. Here. I'm saying it's not doing a great job. Cloud works much better with this, of course. Now we can close all this, take Codex and, for example, write to it, create a project according to the specification and drag this project MD to it. And it will now read this project completely and understand what it needs to do. And it will make us a single-page, so to speak, a single-page online store for a pizza delivery service.
While it's doing all this, I also wanted to show you these things: compile an up-to-date list of MCP and Skills. Skills for coding. Write all this for Codex, suitable for my project. Find an up-to-date list on the internet in English. So now it will go, you have a project, it's already known, and now it will go to the internet and find the entire list of MCP and skills suitable for your project that you are creating. So it suggests Git MCP, F MCP, Memory MCP, and so on. I think it's clear. It suggests skills for you to definitely install Developer, React UI Builder, and so on. I think this is one way to select MCP skills for your project. But in general, you should choose MCP servers yourself. So on MCP Market, choose skills also on MCP Market and in the Skills section specifically for your project, or find ready-made bundles on the internet, i.e., those recommended now for frontend, backend, mobile development, and take the list from there. But ideally, it's better to choose yourself through development, observe from experience how it all works, helps, doesn't help, how it affects your project, and so on. So this is a long iterative process. This would be ideal.
There is also a different type of specification creation. That is, you can simply, well, if you understand, because you need to have an idea of the project you want to create. That is, you can simply turn on the voice recorder, record everything you want to create, and at the end ask it to create a Markdown file in MD format, and it will do something similar. But GPT is not very strong in this. It doesn't do it very beautifully. It's better to ask Cloud to do it. Cloud does it much better. It creates all these Markdown files. Even based on my generation file, it will create a more suitable, correct Markdown in this format, which you can immediately save, download, and use in your projects.
Meanwhile, Codex has finished our entire project. To view it, Codex usually writes what needs to be entered. We open the terminal. I will delete all this now. Look, it says that we should be in the Codex Settings folder. Look, we are already in the Codex Settings folder. And now we need to type npm run dev. We type it. The server has started on its own. We click on localhost:3000 in our terminal, in Codex. Now it will load, and we will see what project Codex 5.3 created based on our documentation. Well, here is such a simple project we got. So we have a certain menu, we can add pizzas. Here is how the cart looks. And we can order all this. Such a simple, ordinary project turned out. To create a more grandiose project with cool animation and so on, you must specify everything in the specification, then we will get cooler projects. Or upload prompts with ready-made projects that you like, so that it, a neural network, creates a new prompt for you based on them.
Also, to see the code of this project, if you are interested, you can just click open. It will open your Visual Studio Code, and you can look at this project. So it will be here. Well, this is more for programmers. For us, coders, it's enough that it just works.
So, this is the video we got. I hope it helped you. And I thank you for watching. Thank you very much. Write comments, leave your opinion about Codex, and also give likes, subscribe to the Telegram channel. And now I say goodbye to you. See you soon. Bye. Yeah.