How Claude Solved AI's Biggest Problem

You probably didn't realize it, but MCP had a big problem. But Claude's new feature, Skills, solves that very problem. Skills is actually a feature that shows us how the future AI landscape is going to look. It's been some time since we realized that there are a lot of repetitive tasks that we do with AI. To solve this, projects were introduced, but they only solved a limited part of the problem. So in this video, I'll take you through what skills are, how you can use them, and to help you solve your repetitive tasks, how you can easily build your own just by prompting. Essentially, skills are just a collection of instructions and resources that we provide to Claude to give it specific capabilities. Instead of functioning as a general purpose agent, Claude becomes a specialized agent with additional capabilities. Each skill that you create inside your project resides in a skills folder in yourclaude folder. Each skill added to your project is identified by the name of its folder. For example, in this project, we have three skills, each serving its own specific purpose. Inside the skills folder, there are several types of files, but the most important one is the skill.md file. At the top of the skill.md file, it has a name and a short description. This helps Claude identify what the skill is called and what it actually does. It serves as a detailed document that references all related files and provides instructions on how to use them. In addition to the skill.md, the folder may also contain reference files that further guide how to use the skill. For instance, in our PDF extractor skill, there is an extraction guide that helps Claude understand how to extract textbased information accurately. These are often supplemented with executable scripts that can be run to perform the specific task the skill is designed for such as the extract text tool that when run basically just extracts all the text from the given PDF. Now let's actually see these skills in action. I'll be using the Claude desktop app to show you how they work. You can view the skills under the capabilities section in the settings. Here you can upload your own or use the ones that are already provided by Claude itself. For example, I can actually use their official MCP builder skills to make new ones. It used the web search tool to access the Figma documentation and after retrieving it, the skill generated the files that can be integrated into our application using the MCP builder skill. It also provided a guide on how to include the MCP server in our project. This is because at the back end, it has the necessary tools and built-in context that is required to build out these MCPs. But these skills are not only limited to developer use cases. Anthropic also added these really creative use cases such as this algorithm art skill which designs a colorful geometric pattern with animations. It first read the documentation, used the appropriate tools, and then generated the animation. This is what the prompt created. You can run it, modify the settings, and experiment with different options to see the results for yourself. Here I used the canvas design skill to create infographics. It basically read the MD files that it needed for the context and also ran a few scripts as its tools. After it was complete, this is what it generated based on the design specifications. There were some problems with it though. For example, the text that was inserted was too small. I did try to generate it again, but it didn't make much difference. There must have been some problems in the tools of the skill. Other than that, Claude also introduced specific skills related to documents such as Word and PDF files, Excel sheets, or even PowerPoint presentations. I tried creating a PowerPoint and saw that it basically converted simple HTML pages into a PowerPoint. While the design that it made here was too simple, I feel like with better prompts or even tweaking the skill a little bit would improve the output. Before we continue with the video, here's a word from our sponsor, Scalekit. The complete authentication stack for AI applications. If you're building B2B AI products, ScaleKit gives you everything you need to manage identity for both humans and agents. It lets you authenticate AI agents, add a drop-in OOTH layer for MCP servers and APIs, and secure users with passwordless flows like magic links and pass keys. Plus, you can go enterprisegrade with SSO and SCIM support. What makes Scale Kit stand out is its modularity. You don't have to replace your existing O setup, whether you're on Ozero, Superbase, or a custom open-source solution. Scale Kit seamlessly layers on top to handle advanced enterprise and agent identity needs. In short, it's built to help AI startups scale securely and effortlessly without rearchitecting your authentication from scratch. Scalekit authentication built for the new generation of AI native applications. Click the link in the description and start building today. So now you might be wondering if MCP already exists, why should we use Claude skills? To answer that, we first need to understand the primary difference between MCP and Claude skills. Even though Claude skills and MCP might seem like they overlap, they actually don't. To put it simply, MCP is about adding more capabilities to Claude that it doesn't already have by giving it access to external tools. On the other hand, Claude skills are designed to enhance internal functionality. Skills contain prompts, instructions, data, or scripts grouped in a folder to help Claude handle specialized tasks efficiently. This reduces setup time and ensures consistency in repeated tasks. But before these skills, people used to implement these internal tools with MCPs as well. But skills solve one of the biggest problems that those MCPS had. While MCP exposes all the tools and their descriptions directly to the model in the context window, greatly expanding the context window, skills handle this more efficiently. Instead of including a large prompt with many lines from the file, skills only expose the initial name and description to the model, allowing it to identify which skill it has. The description is a oneline summary that briefly explains the functionality, and this is all it needs in its context window. This is how Claude knows what skill it has and how to use them. And whenever it detects the need to use a certain skill, it can refer to its documentation for further details. But Claude could have made these skills so much more powerful. However, they implemented a limitation which turned out to be its biggest flaw. Claude skills cannot use the network. This means that they can't send any API requests. This means that you cannot make a Figma skill and use it as the Figma MCP because it uses the Figma API to get the design data. However, there is a workaround. The skill can use the web search tool from either claw desktop or clawed code to fetch the data it wants from the web, but it still won't be able to request data as the Figma MCP does. By now, you've probably realized how much repetitive work you can automate with these skills, and you're probably thinking if you would need another course to learn how to make new skills. But Anthropic offered a really clever solution to that by adding in a skill that actually builds skills. You can use the skill builder to build out skills for any use case right in Claude. You just tell it what you need and based on your requirements it may ask you further questions and after that it just builds out your skills including all the markdown documentation and the tool scripts as well. But this is only available in the claude desktop app. If you want to use it with claude code then you'll have to get the source code of the skill builder as they launch skills. Anthropic also released a repository that contains many commonly used skills and in that you'll find the skill creator as well. We'll also link it down in the description below. Now to get it inside Claude code, you need to download the skills and add them to the Claude folder in your project. After that, Claude will automatically recognize that it now has a skill called skill creator. Now using the skill creator, I'll ask Claude code to create a new skill that specifically optimizes a website for SEO. After Claude was done, it had generated a new skill that included all the necessary tools and scripts required for it to function effectively. I also went ahead and created these three skills, each with its own assets and skill file. To test out the SEO skill, I used it on a pre-made site that I was building. It went ahead and analyzed the website's current performance and then applied a few improvements. In the end, the overall performance score increased from 86 to 94. This shows how you can use different types of skills for various use cases for your application. You've probably seen the new chat GPT apps and widgets which allow you to get responses in these beautiful UI formats. With skills, you can unlock another useful capability of Claude, the ability to use visualizations in your local environment for various use cases. For example, if you want to analyze your data visually, using these skills with Claude will be very helpful. With these Claude skills, you can use visualizations in your own local environment just like in chat GPT. It's different from chat GPT because chat GPT can use API calls and HTTP requests to communicate over the internet while Claude's environment is fully local and cannot access the internet. However, we can still use these skills to create beautiful visualizations of anything. For example, I created a skill that uses a consistent HTML template to render a visualization for the search data. So essentially, you can think of it as widgets inside claude code. I gave it a prompt to visualize the top 10 trending AI tools right now and it ran several scripts and web searches. In the end, we got a really beautiful visualization. It showed me the highest ranking AI tools currently available. The animation was also very smooth, and now whenever I ask for new AI tools, I'll get them displayed in this same widget format every time. That brings us to the end of this video. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below. As always, thank you for watching and I'll see you in the next