How I set up Claude Code for faster and better AI code

Key Takeaways

Clot code has already become a standard for coding software projects but still many people are simply writing English prompting those agents to build applications for them and they're leaving a lot on the table. Cloud code is much more powerful. In this video I'm going to be sharing a selected few best practices and tips and tricks so that you get the most out of clot code. This thing is insanely powerful. So for those of you who do not know, plot code is a very very useful very capable command line tool for agentic coding. Now you need to understand how to set it up in your machine, how to structure your software projects, what sort of commands to use, how to create custom commands, how to leverage sub agents so you can spawn multiple cloud code agents to do different types of tasks for you. how you can invoke multiple uh cloud code instances to work on different features leveraging get how to connect your software with external applications using MCP servers. So how to add MCP servers and what sort of MCP servers uh are the most effective for coding software projects. I'll be sharing everything in this video. So let's get started. All right. So the first step is to install clot code. I already have it installed. You can go to the quick start guide, copy this command, npm install and paste it in your terminal. I'm using a terminal called warp which again is very powerful. So I would recommend that you check this out. They also have their own agents but I like to use clot code within warp. All right, I am here. All you have to do is paste the command, hit enter, and it'll install clot code for you. Now all you have to do is uh navigate to your project directory, your software project directory and simply write claude there and there you go. I simply will have to first of all login. So there are like a bunch of commands that you can use. Yes, proceed. It'll ask you allow full access. It's okay. And now they have already given you some tips. Run slashinit. These are these slash commands are part of uh clot here. By the way, I'm using clot code in a terminal. Another way to use clot code is in your code editor. So this is cursor. I if I want to use uh clot code within my editor, I can do that as well. I can open up the terminal. If you're using VS Code, Windsor, whatever editor that you're using, you can open up your terminal. And again here also you can type in like claude and it will start the claude session for you. Many of you would want to use claude code uh like uh the cursor terminal the cursor chat that we used to get. So this was the new chat feature that uh uh we had in cursor. If you want to leverage clot code in a similar interface you can do that as well. All you have to do is just uh uh press command shiftp and here you will have to install cursor command this uh click on this and it will so all we want is this clot code over here should be connected with our editor. So whenever I want to make some changes, let's say if I am in a particular file, the CSS file and if I want to make some changes uh in this like selected uh lines. So I want this editor to be connected to my clot code and in order to do that all you have to do is delete this session and you can press command escape and it will open up slot code in a similar fashion as a cursor chat. So here now you can just start chatting with your project like you were doing in your AI code editor. Now the first thing to note here is that cloud has got multiple modes. So here if I press shift tab you can see auto accept edits on. That means whenever claude has reached a stage where it needs a user input this would be turned on. For some of the sensitive commands it'll still ask you it'll beep. uh you can set up uh your terminal in that way. I'll show you how to do that as well. And with first shift tab, auto accept edits on. And if I press it again, shift tab. Now this time it has got plan mode just to plan everything out. There going not going to be uh any code changes that cloud code would make. So plan mode is good. So you plan out your entire execution, your architecture, uh the tools to use, everything, the research etc. All those things can be done in plan mode. And once you're sure, okay, yeah, this is everything is good, then you move on to the execution with auto accept on. I usually run it on auto except on. Now claude has got a bunch of slash commands that we can use. Here also they've mentioned in the tips that you run /init to create claw.md file with instructions for claude. Now these are sort of like system instructions for claw code agent. And here if I simply do slashinit. So you can see there like a lot of these commands that have been created by clot code for us that we can you know start using. But slashinit what it does if we hit enter it initializes this clot.md file where the entire structure of this project. So if you are in your software project directory you keep running this /init command. you know at regular intervals it'll keep updating the claw.md file in your project. So this will pull together you can see it is reading searching it is analyzing my complete project. This project is a chrome extension and uh this allows me to chat with any web page out there. So let's see what it does. It'll set up. So I'll speed this up a little. All right. So there you go. You have the opportunity to edit it as well. Would you like to proceed? It is asking me yes. Okay. So this will create a claw.md file for me in my project. And there we go. So we have got the claw.md file. You can see it has got the structure everything with instructions. The claw.md is a special file that contains all the instructions you know the bash commands that we want to use the project structure. So all of those important details that you want that your agent should always have in context whenever it is starting a conversation. So this is basically that go-to file that will always remain in context for your agent to take any action. Context is the most important thing when working with clot code and uh clot code basically leverages three types of memories. So you need to understand uh where to put this clot MD file. So one is uh project memory. So dot /cloud.md. So this is like uh in my projects. So here in this particular case, whosoever is working on this chrome extension in my team, everyone would follow this claw.md file which is the project instructions. So this is my project based memory. If I add if I add this claw.md in a hiddencloud folder in my home directory then that becomes my global instruction. So this is like personal preference for all the projects and clot code refers to every clot.md file uh you know it recursively refers to this system instructions across all my machine and then makes decisions across all the projects. So if you are working on let's say 10 different projects and you want all those 10 projects to have some sort of consistent behavior that say you know it should have the sort of like security uh checks across all of those projects. So you can have those types of instructions in your home directory in your user memory. Another very useful feature in clot code is custom slashcomands so that I don't have to write frequently used prompts and guardrails whenever I want clot code to do something for me. So these will fetch these frequently used prompts as markdown files that clot code can then execute for me. And the syntax is pretty simple slash command name and then the arguments. Here you can also provide optional arguments. you can pass it on to the command and then command would kind of like format those variables, name of files, it could be anything uh into your prompt itself. So a bunch of command types. So let's say I want to create this optimize command. All I have to do is go into my folder and you can see they are asking you to create a folder. This is a hidden folder. Cloud and within this I have commands directory. within the commands directory, I need to put my markdown file for that uh frequently used prompt that will then become my command as well. So the command name is nothing but the markdown file name. So here they're asking us to create this directory. Now one thing is this is a bash command. This is a terminal command that I want to run but I am currently in my cloud code instance. So if you want to turn this into bash mode, simply press exclamation. As soon as you type exclamation, it turns on the bash mode for us. And now I can simply paste my command. Hit run. And you can see I have claude command ready. Okay. So this directory is created for us. Now all I need to do is create a new file. I'll call it optimize.md if I'm creating an optimize command within this particular project. And in this all I need to do is simply specify my prompt. I can also do this uh in a terminal but currently I'm just specifying it here in the file through the editor. Analyze this code for performance issues and suggest optimizations. That's it. So now this optimize command is available for me. So if I simply type optimize here and press enter, cloud code would start thinking about what to do in this project. So here you go. I've already completed a comprehensive optimization for your codebase. Great. Another type of command is your personal commands. You want to do something for example security review. So review this code for security vulnerabilities. So this would be security review.md. So whatever your command name, so this command name is also going to be the name of your markdown file. And one other thing that is a very good pro tip for you, if let's say you have front end, backend, different types of uh sub directories in your main project. So you can also create like commands and then /frontend. So you can create another subdirectory which is called front end or back end and then create a project specific custom slash commands. So if I want to you know create a new component for a particular thing /component this is how I can do that. If you want to pass dynamic arguments your command your prompt should have this dollar argument placeholder. This dollar argument placeholder sort of works like your Python formatting. So all you have to do is slash your command. This is the file fix issue.m MD and then here whatever I write ahead of that command that will be formatted into my prompt uh directly. Now one very useful custom /ash command that I have come across is this explore plan code test.md. So everybody keeps uh suggesting you that you should first explode then plan then code and then test it all out. So here's this prompt. All you got to do is copy this entire prompt. I'll share the gist uh and the entire prompt uh in the description. You'll find the link in the description. All you have to do is create a new file. Uh let's call it explore plan develop text or what is it? Explore plan code. So I'll rename this rename this file explore plan code and then test MD. So this becomes my command epect and I'll paste the entire prompt here. So you can see at the end I've got dollar arguments. So all I need to do is /ct and then I'll specify you know what do I want to explore, what do I want to plan. So I can basically leverage this entire thing. So explore plan code test workflow. uh when you start doing some projects. So whenever you are starting off a project, this command, this prompt would be very very valuable. This is something that I have come across. So yeah, do utilize this. This is from Sergey uh yeah on GitHub. A very underrated feature that cloud code provides you is sub agents. So you can ask clot code to spawn multiple agents and each sub agent then will solve a particular problem. For example, I can create a deep research task that will go through chain, fire crawl, and screenshots API to add to my Chrome extension project. And you can explicitly tell cloud code you should spawn multiple sub agents for this research task. That's it. And all I'll do is hit enter. Now you would see that it will spawn up multiple tasks, multiple sub aents for me to get this job done. So you can see research lang chain integration. So first of all the plan is ready uh with me. There you go. So you can see the first task research lang chain integration. This is second task. So these two are running parallelly. This is like parallel processing that's happening currently. Research screenshots API fire call advanced features chrome extension AI architecture. So there like five uh four sub aents that are running in parallel and getting this job done for me. So one very good use case for this is research for me. Uh whenever I find myself using sub agents uh it's probably going to be like a breadth first uh task for me and research is one very good use case for that. So if I want to do like multiple uh you know researches uh at the same time I leverage this feature. Okay. This one is a personal favorite. Uh this is whisper flow. Now I know that many a times I would get uh tired of typing these prompts. So all I want to do is press some hotkey on my keyboard and uh just voice it all out. So I want to dictate my prompt and that prompt should be very very good. It should be wellcraftrafted prompt. So I use this uh app called whisper flow and I don't type, I just speak. This is one very good workflow that I found for myself. It takes like you know a couple of minutes to get used to and after that you just remind yourself to just simply uh voice it all out and this is AI powered so it autoedits it removes all yours and h and all the fellow words and simply write a wellcrafted prompt for you directly here for example if I simply press my hotkey here and write hey create a deep research to study langen document ments, fire crawl documents and screenshots API to integrate into my product. There you go. So, it has basically written that prompt for me from that dictation. It's it's a Mac app that I have installed and uh it g they give you like 14-day free trial and after that uh you have to pay like 15 bucks uh or something like that. Uh and and this the grammar etc. everything will be taken care of. So pretty good tool that I've uh integrated into my workflow. Now comes the most important part, the most important feature which is how you can connect your claude code with external tools and model context protocol MCP comes in very handy here. You have to keep in mind that claude code has access to all the other terminal commands. So if you are let's say using GitHub CLI using the GH command. So if it is already installed on your machine, so cloud code is smart enough to leverage that SDK for you. But there are some tools for which command line interface SDKs are not available. So for that you can connect and your MCP servers you can configure those remote MCP servers. Uh there are bunch of different ways that you can do that. If you want to connect any of your local servers, so there's a way to do that uh that they've provided. Uh if you want to leverage any server send event, uh there's a way to do that. The most common workflow is adding a remote server, so an HTTP server that you want to add. So this is the command for that. If you want to check how many MCP servers your claude code is connected with, so all you have to do is claude MCP list in the terminal. just specify this. Currently I have this context 7 MCP connected to my clot code. And here also you have to keep in mind that you can uh have different MCP servers based on uh the project requirements. So I can create uh dot MCP.json file and this will contain all of this JSON data. So here you can see if I want something they have specified project scope servers enable team collaboration by storing configuration in MCB.json file at your project's root directory. So anything that I want to do now anything I want to add any particular MCP server it'll get added directly to this particular file. So all I need to do in order to add any new uh MCP server for example let's look at context 7. Context 7 for those of you who do not know it provides up-to-date documentation for your application. So if you are using let's say langchain langraph and you want to uh you're using a particular version of langraph. So they have version control documentation of all of these different packages that are out there. So it's pretty good. Uh I think this is one uh MCP server that I would recommend to all of you. And in order to add this uh you can check context 7 GitHub and if you go to the GitHub server here you would find a command. So they've given commands to install it in different tools. Here you can see the one that we want to use is this remote server uh claude MCP ad. Just copy this command paste it here. cloud mcp add d-t transport http context 7 the name of my mcp and then the remote link that's it all I need to do is do uh pass this command run this command and it will configure context 7 with clot code and uh at any point of time you want to get a particular [Music] mcp so you can specify cloud mcp get and then you can specify the name of that mcp it'll give you all of the details scope local type http URL this and to remove the server you can run claude mcp remove context 7 minus s local now there are bunch of these flags that they have provided use the minus s or d- scope flag to specify where the configuration is stored do you want to store it locally default uh project structure mcbpjson file or user that means available across all of my projects that is the global configuration. Now you must have seen people running 10 different cloud code sessions and I don't know what they're working on. I personally think it's an overkill you should be using three to four maximum cloud code sessions to work on independent features and the best way to do that is uh using work trees. So they allow you to check out multiple branches from the same repository into separate directories. So each work tree will have its own working directory with isolated files so that there is no conflict and they'll all share the same get history. Now in order to uh leverage this feature first of all you have this command get work tree add uh and then it creates a folder in the root directory. It'll come out of the project folder create a folder outside of it and then create a new branch which is called feature A. So here uh I'm just calling it feature A. Uh let's copy this command. This is my same Chrome extension project test claude code. And I'll paste the command get workree add. It'll create a new folder which is called project feature A with a new branch feature A. I'll hit enter. Now if I come out I'll see this project feature A. I'll go inside project feature A. I've got a feature A branch. Now within this I can start a claude session. Great. Similarly, if I want to work on another feature simultaneously, let's say project feature B. I will simply copy the same command with the so get work tree add this is going to be let's say project feature B and uh I'm going to create a separate directory for this as well. That's it. And all I need to do is project feature B. There you go. We have feature B command here. And I'll launch a claude instance here as well. And that's it. Now this claude session is going to work on feature B. This one is going to work on feature A. You can also think of front end, backend, infra, and then all of those uh sessions combining together and working on your project. So that's how I personally see the use case for running multiple cloud code sessions. Now let me share some really quick commands and hacks that uh I have developed and learned to use over time. So first one is let's say you have uh prompted uh cloud code to do something and you hit enter but suddenly you realize okay that's wrong. So I press escape to interrupt and uh end that command and now I can just rewrite my prompt here. The best thing about using plot code in an editor is uh that I can select a particular function number of lines. So you can see 18 lines selected and now I can prompt uh how to make changes to these 18 selected lines. So this is like connecting my IDE and uh the context of these lines are already provided to cloud code. You can copy the screenshots of your application that you have developed to fix let's say any UI issue. UX issues. So all you need to do is take let's say this screenshot. This is the Chrome extension that we are building. And I can just simply drag and drop it in my terminal. The image will get pasted like this. And then I can just prompt uh make the UI look more elegant. Or if you have let's say any Figma screenshots, you can just copy those things. Copy the screenshot, copy the image and paste it here and it will start working for you. Now, if you don't want the clot code to hallucinate and you have a really, really long chat that's been working on a feature and now you want to work on a new feature, I would recommend that you use the clear command, clear conversation history, and free up the context. This will really help clot code deliver highquality output and high quality code for your features. So this is very very important. Keep running/clear and/init commands regularly in your project. One rookie mistake that I was making early on was to asking clot code to start building my application. But uh you should always ask clot code to let's say if you want it to go through a particular URL if you want to create let's say uh you know a deep research task. So always ask clot code to first do that research find out the state-of-the-art tools the best ways updated documentation for the frameworks that you're trying to use and then uh create a plan and execute. Now if you're wondering that there are a lot of things that you have to keep in mind there are a lot of tips and tricks. So there are some third party packages also that uh you can leverage this is called super cloud v3. Now they are extending the capabilities of cloud code with specialized commands. They have created some custom commands for you some personas that they have defined and uh some MCP server integrations. So this is the version three. It's pretty simple to install and uh here they focus on 16 essential commands. So these are 16 custom commands uh that you can then start using uh the persona architect front end backend analyzer. They have created nine such personas for you to work with. Then we have MCP integrations. They provide context sequential magic playright. All of these will be installed directly uh for your plot code to start using. Uh installation is pretty simple. clone the repo, go to the folder and run this particular command. That's it. So I am going to do a dedicated video on all the other types of software projects and packages that extend the capabilities of clot code. But uh these things are also available for you to directly start leveraging these best practices.