Claude Skills & Plugins Tutorial: How to Use Claude AI Like a Pro (Beginner's Guide)

0:00 Claude is one of the most powerful AI tools, but Anthropic made it possible to make it even more powerful for your own needs. You can teach Claude new skills, give it access to your data and third-party services, or install ready-made solutions in just a couple of clicks. Let's break down how all of this

0:17 works. Most of you spend the majority of your Claude time right here, Claude chat, and this is where the ecosystem is most accessible. Everything we're covering today is reachable from the left sidebar or from your profile settings. So, instead of treating Claude like a single chat box, I want you to

0:34 start seeing it as a full workspace. The chat is just the surface. Behind it, you have settings, skills, connectors, plugins, and the MCP layer that lets Claude interact with tools and data outside the conversation. So, let's start here, inside Claude chat, and work our way through each layer one by one.

0:54 Skills are the most misunderstood piece of this ecosystem. You're not reprompting from scratch every time. You're teaching Claude once, and it remembers the way you want a specific task handled. You'll find them under customize, then skills. Hit add to open the creation form. You give the skill a

1:11 name, a short description so Claude knows when to activate it, and then the actual instruction. That description matters more than people think. Write it like a clear use case. When I'm researching a video topic, when I'm reviewing a script, when I'm analyzing a client brief. Too vague, and Claude

1:27 won't know when to activate it. Too narrow, and you'll trigger it manually every time. Now, let me show you two skills I actually use in my workflow. The first one is my video research skill. Whenever I'm working on a new video topic, I want Claude to structure its research output in a specific

1:43 format: main claims, source types flagged, gaps identified, competing angles listed, and raw facts separated from interpretation. Without a skill, I paste that instruction every single time. And realistically, after doing that 10 or 20 times, you either forget part of the instruction or you make it

2:02 shorter or you get lazy and accept a weaker research output. With a skill, Claude just does it consistently, every time, without me rebuilding the instruction from memory. My second skill is a script reviewer. I trained it to check pacing, sentence length for voiceover, whether the hook lands in the

2:20 first 10 seconds, and whether any section explains too much without moving the viewer forward. These two skills alone save me about 30 minutes per video and more importantly, they make the output consistent. I'm not relying on memory. I'm not rebuilding the same instructions. I'm turning repeatable

2:36 judgment into a reusable system. And you don't have to build every skill from scratch. Hit browse [music] and you'll see a catalog of skills already published by Anthropic. Pick something that maps to what you do and it's installed with a click. Start with one workflow you repeat weekly and turn it

2:52 into a skill. Do not start with something abstract. Start with something annoying, repetitive, and specific. Research format, script review, email tone, meeting summaries, client brief analysis, whatever you do over and over again. Everything we just covered with Claude becomes much more powerful when

3:10 it is all in one place. Inside AI master, you can work with Claude, Gemini, and the other latest top models all in one window. So, instead of paying for multiple overlapping subscriptions, jumping between tabs, and rebuilding context every time, you choose the right model for the task, run it inside one

3:28 workspace, and pay by token cost, which comes out cheaper than stacking separate subscriptions. You can even run the same prompt across different models and compare the results side-by-side. Memory, web search, and file attachments are built in. For example, I can use Claude to structure a script, another

3:46 model to generate alternative angles, and another one to refine the final wording, all without leaving the workspace. And AMS goes beyond chat. The platform also gives you image generation, voice over, and video generation in the same place. You can create consistent AI characters that

4:04 keep the same look across generations. Publish and monetize those characters and share generated content directly through the platform. That same LLM core powers content agents. They help plan your channel, draft speech-ready scripts, and design thumbnails. The full agent crew is still rolling out feature

4:22 by feature. There's also a full academy inside, more than 200 lessons, around 30 hours total, taking you from beginner to expert in AI content production. So, this is not one random AI tool. It's a full production environment for people who create content seriously. And there is a real community behind it. More than

4:41 12,000 users are already building with these tools, sharing workflows, testing ideas, and creating content with AI. On the side, you can also see real testimonials from creators using the platform on real projects. There is an annual plan with a strong discount and a 7-day money-back guarantee, so you can

5:00 try the platform without much risk. Once you're in, the first thing I recommend is opening the onboarding agent and describing your project, your channel, niche, audience, tone, and goals. That gives the system the context it needs, so you're not starting from zero every time. All the links are in the

5:16 description. Before we get to plugins, two things people constantly mislabel as plugins, web search and code execution. These are not plugins, they're native capabilities built directly into Claude. No installation, no directory, no configuration. You access them from the tools menu next to the chat input field,

5:34 or Claude activates them automatically when it decides the task needs them. Web search is the one I use every time I'm researching something that moves fast, like anything in AI, a new model, a new feature, a pricing update. I don't want Claude relying on memory or outdated info. I want it to check the current

5:52 source and reason from there. And the best part, I don't have to remember to enable it. If the question is about something recent, Claude just searches. Code execution works the same way. Claude decides when the task needs it, spins up a sandbox, runs the analysis, and returns the result. For

6:08 non-developers, this is probably the single most powerful capability jump you'll get. You upload data, ask a question about it, and Claude actually runs the numbers, writes the code, executes it, and gives you the result right in the chat. And for developers, it removes a lot of small manual work.

6:26 Verdict, these are already working out of the box. You don't need to enable anything, just start asking Claude questions that would benefit from a search or script, and it handles the rest. Now, the actual plugins. The plugin is a ready-made bundle, skills plus connectors packaged together, and

6:44 sometimes sub agents and hooks. Though, those only activate in co-work. In regular chat, they simply stay inactive. The point is, instead of configuring each component by hand, you install the whole bundle in one click, and it works out of the box. Plugins are available on all paid plans. You find them in the

7:01 plugin directory under customize plugins, then browse, or at claude.com/pigs in your browser. Open the plugin card, and you'll see what's inside, the skills it adds, the connectors it brings, what it's designed for. First, enterprise search. It connects Claude to your email, chat, documents, and wikis, so

7:20 you can search across every tool your team uses from one place. Instead of jumping between Gmail, Slack, Drive, and Notion looking for the same reference, Claude has one entry point into all of them. Second, data. It's built for working with SQL, data sets, and analytics. If your work touches numbers

7:39 at all, dashboards, reports, exploring raw tables, this plugin gives Claude the skills to help without you writing queries by hand. Now, an honest note, the plugin directory right now is heavily skewed towards specific roles: marketing, sales, developer tools, data. If a plugin doesn't match what you

7:57 actually do, don't install it in bulk. Pick precisely. The good plugin should either give Claude access to something it could not access before or let it perform an action it could not perform before. If it does not do one of those two things, be skeptical. Verdict: browse the directory once, install one

8:16 or two plugins that map to a workflow you already have, and skip the rest for now. So, that's what MCPs actually unlock. Here's one I've been running for weeks. More and more buying decisions start inside ChatGPT, Claude, Gemini, and other AI assistants. The problem is most companies have no idea whether

8:33 they're actually being recommended or why not. That's what Zero Rank solves. It tracks how your brand shows up across the major AI platforms, who gets recommended instead of you, and what's driving those results. And it ships with a Claude MCP that plugs straight into your workflow. With the MCP connected, I

8:52 can just ask Claude things like, "Am I showing up in AI answers? How can I increase my visibility and brand mentions?" Or even, "Give me a content plan for the next month." The MCP is the conversational layer, but the real work lives inside the platform. You can track custom prompts, monitor competitors, and

9:10 see the exact citations behind every AI answer. Zero Rank doesn't just report on visibility. It hands you prioritized fixes and helps you optimize the content around the gaps. Try Zero Rank for free. The Claude MCP is included with every plan, and the link is in the description. This is the part most

9:30 people completely skip, and it's probably the biggest productivity unlock in this entire ecosystem. Connectors give Claude read and write access to the services you already use every day. Gmail, Google Drive, Google Calendar, Slack, HubSpot. These aren't hypothetical integrations. They're right

9:48 in the panel, ready to connect. And beyond the built-in list, there's a full directory of MCP-based connectors you can add on top. And the reason this matters is simple. Most of the useful context in your life is not inside a blank chat window. It is in your inbox. It is in your documents. in your project

10:06 folders. It is in your notes. It is in your team conversations. If Claude cannot access that context, you have to copy and paste everything manually. That works for one small task. It breaks when you are trying to use AI as a serious workflow tool. Once a connector is set up, it just sits there quietly, ready to

10:25 go. That checkmark next to Gmail, that means Claude can now read your inbox and act on it whenever the task requires. You can ask Claude to summarize a long client thread and tell you what actually requires a decision. You can ask it to draft a reply that matches the tone of your previous messages, not a generic

10:45 template, but something that sounds like you. You can ask it to scan the last 2 weeks of emails and pull out anything you promised but haven't delivered. The key difference: with no context, a model gives you polished but generic writing. With inbox access, it works from your actual conversations. It knows what was

11:03 promised, who's waiting on what, and what tone you use with each person. That's when a chat tool becomes a real assistant. Google Drive is the second one I'd add. Drive is where research lives, where docs live, where the messy middle of most projects lives. Once Claude can read your Drive, you can

11:22 point it at a folder of notes and ask for a structured synthesis. You can pull data from a spreadsheet without opening it. You can draft a new document using context from three existing ones without copy-pasting anything. [music] The old workflow, open every doc, skim it, copy the important sections, paste them into

11:41 Claude, explain what each document is, then ask for synthesis. The connected workflow, point Claude at the files, describe the outcome you want, and let it work from the source material directly. And the value isn't just speed, it's context integrity. Claude isn't working from a rough summary you

11:58 pasted in, it's working from the actual files. Nothing gets lost in the translation. Setup is the same for every connector. You find the service you want, hit connect, and go through a standard OAuth authorization. You're giving Claude permission to read and, in some cases, write on your behalf. Takes

12:15 maybe 2 minutes per service. Pay attention to permissions. You're connecting real accounts with real data, so don't treat this like a random browser extension. But once it's connected, the workflow change is immediate. And here's the part most people miss. Beyond the built-in connectors, there's a full directory of

12:33 MCP-based integrations you can browse and install. Slack, Notion, HubSpot, databases, dev tools, even GitHub if you work with code and want Claude to review pull requests, explain functions in the context of a repo, or summarize commit history. The catalog is growing every week. Anything that used to require

12:53 copy-paste is quietly becoming a one-click connection. Verdict, connect Gmail and Drive today. Then open the browse catalog once and add whatever else maps to work you already do. These integrations alone will change how you actually work with Claude on a daily basis. Okay, so we've talked about

13:12 skills, plugins, and connectors. All three of them, at a technical level, can run on top of the Model Context Protocol, MCP. This is Anthropic's open standard for how Claude connects to external tools and data sources. You don't need to understand MCP deeply to use the features I just showed you, but

13:31 you do need to know it exists because it's the reason this ecosystem is extendable beyond what Anthropic ships natively. In other words, MCP is the layer that turns Claude from a model that talks into a model that can work with tools. Think of MCP like a USB-C port for AI applications. Instead of

13:51 every tool needing a custom one-off integration with Claude, a tool can expose an MCP server and Claude can communicate with it through a standard protocol. That's why this matters. It is not just another feature in settings. It is the foundation for a much larger ecosystem of integrations. In the Claude

14:09 chat, you'll mostly experience MCP through the connectors you set up. Those are MCP-powered under the hood. The important thing to know at this level is that any service that builds an MCP server can be connected to Claude, which means the list of possible integrations is essentially unlimited and growing

14:27 every week. There are already hundreds of community-built MCP servers. Some connect Claude to niche internal tools, some to databases, some to APIs that don't [music] have official connectors yet. If you ever hit a wall where Claude needs access to something that isn't in the official connector list, MCP is your

14:45 path forward. For example, maybe your company has an internal database, or a custom CRM, or a private analytics system, or a specialized API that would never get a native connector because it is too niche. That is where MCP becomes important. You do not have to wait for Anthropic to build every integration

15:05 manually. If the tool can expose an MCP server, Claude can potentially work with it. Now, for most people, the safe and simple entry point is still connectors in Claude chat. But if you work with developers or you are a developer yourself, MCP is where the ecosystem really opens up. We'll go deeper on this

15:24 in the Claude Code section. Everything I showed you in Claude Chat lives inside the browser. Claude Co-work is where those same ideas get amplified through a dedicated desktop workspace built for real ongoing work. I'm not going to rehash the full chat setup here. What I want to show you is two capabilities

15:42 that only really make sense in Co-work and that completely changed how I think about the Claude ecosystem because Co-work is in Claude, but bigger. It's Claude with an entirely different layer of automation on top. Role-specific plugins, sub agents, and native connectors that come pre-configured,

16:01 ready to go. That changes what a single install can do. In Claude Chat, I built two skills from scratch, my research skill and my script reviewer. That took time and it works for me because I know exactly what I want. But in Co-work, Anthropic ships plugins built for specific roles, marketing, sales,

16:20 finance, legal, data, product management. Each one is a complete workflow package. You install it in one click and Claude instantly understands how to work like someone in that role. Let me show you the one I actually use, the marketing plugin. Look at what's inside. First, the Try Asking section, a

16:39 set of ready-made prompts Claude is tuned to handle out of the box. Draft a blog post with SEO optimization, plan a multi-channel campaign, review content against brand voice, analyze competitors, build a cross-channel performance report, audit SEO, and find content gaps. Then, look at the skills.

16:57 Eight {slash} commands {slash} competitive brief {slash} brand review {slash} content creation {slash} campaign plan {slash} performance report {slash} SEO audit, and more. Each one is a full workflow, not just a prompt. Here's what I mean. I open a new task in Co-work, type a {slash}, and I get a

17:17 full menu of every workflow across all my installed plugins. Hover over {slash} competitive beef and Claude tells me exactly what it does. Research competitors, generate positioning, spot content gaps, identify threats. I click and here's where it clicks for me. Claude doesn't dump a wall of text at

17:36 me. It opens a structured form. Which competitors? What does my company sell? Any differentiators I already want to highlight. That is a huge shift from chat. In chat, I had to remember what to prompt, what format to ask for, what edge cases to include every single time. In Cowork, the plugin already knows. The

17:56 form enforces structure. The workflow enforces quality. And I get consistent output every time without rebuilding the prompt from memory. That's what I mean when I say Cowork amplifies chat instead of replacing it. Verdict: If any of Anthropic's role-based plugins map to what you actually do, marketing, sales,

18:17 finance, whatever, install it and use it for a week. It's a shortcut to workflows you would otherwise spend hours building yourself. Second capability, and this one I did not see coming, the productivity plugin. Look at the skills section, {slash} task management {slash} memory management {slash} start {slash}

18:36 update. Compact set, but each one is designed to run across your entire productivity stack. Now, look at the connector section. Nine connectors in a single plugin install. Slack, Notion, Asana, Linear, Atlassian, Monday, ClickUp, Google Calendar, and more. In Claude chat, I have to connect Gmail,

18:58 Drive, and each other service one by one. Each OAuth flow, each account, each permission screen. In Cowork, one plugin brings the entire productivity stack with it. And it's not just the connections, it's the workflows on top of them. Claude uses these skills across all the connected tools automatically.

19:16 You ask Claude to catch you up on stale tasks and it pulls from every connected service, cross-references what's active, what's stalled, what's blocking others, and gives you a prioritized summary. You don't tell it which sources to check. You don't tell it what stale means. The plugin knows. That is a level of

19:35 automation that doesn't exist in chat because chat wasn't built for this kind of cross- tool orchestration. Co-work was. Verdict, install the productivity plugin even if you already use one of the tools it connects to. The value isn't in the individual connection, it's in Claude being able to reason across

19:54 all of them at once. I did a full breakdown of Claude Co-work in a separate video. I'll link it at the end. But these two plugins alone are the argument for moving from chat to Co-work if the ecosystem starts to feel limiting. All right, Claude Code. I'm not going to do a full breakdown here. I

20:10 already published a complete guide to Claude Code on this channel and I'll link it at the end. What I want to show you is one thing, how the same ecosystem, skills, plugins, connectors, MCP, extends into development because this is where Claude stops being an assistant and starts being

20:28 infrastructure. This is the Claude Code interface. You get a coding focused workspace, sessions, artifacts, statistics, a model of your own usage, where you spend tokens, when you're active, which model you rely on most. But that's not what I want to focus on. I want to show you how the ecosystem you

20:48 already learned about, skills and plugins, carries directly into code. Same customize menu, same skills panel, and this one, skill creator, comes installed by default from Anthropic. Now stop and think about what this is. This is a skill that helps you create new skills. The ecosystem is building

21:09 itself. Instead of writing a skill.md by hand, trying to remember the structure, the metadata, the trigger conditions, you ask skill creator to help you build one. It walks you through the process, asks the right questions, and produces a working skill you can use across chat, co-work, and code. That is the

21:28 definition of a self-extending system. Now look at plugins. Same plugins as co-work, same directory, one-click installs, marketing, productivity, data, legal, sales. The exact same ecosystem lives in chat, co-work, and code. You install a plugin once, it works everywhere it makes sense. That is what

21:50 I mean by ecosystem instead of features. And here's where it gets bigger. Beyond the official Anthropic plugins, there is a community catalog at claude.com/bangs. Dozens of community-built plugins and MCP servers, connectors for databases, developer tools, project trackers, productivity apps, communication

22:11 platforms, anything you can imagine connecting Claude to, someone probably built the server for it. That's what MCP unlocks at the developer level. You are not limited to what Anthropic ships. You are not limited to what the community ships, either, because if you can build an MCP server, you can plug Claude into

22:30 any system you already use. Local database, internal API, proprietary data source. If it can expose an MCP endpoint, Claude can reason over it. For anyone who works with code, even just reviewing it, not writing it, spend an afternoon in this directory, pick one plugin that maps to a system you already

22:49 use, install it, run a real workflow through it. That is how you find out where MCP is actually useful for you, not by reading about the protocol, but by using it on something that already exists in your work. Verdict, if you write code or work with developers, Claude code plus the community MCP

23:10 catalog is where this ecosystem stops being a chat tool and starts being real infrastructure. For everyone else, the plugins and connectors in chat and code work are your entry point to the same underlying protocol, just with guardrails on. Okay, let me close with a short list I promised. Here's what I'd

23:28 actually set up this week if I were starting from scratch with Claude Pro. First, build two skills before you do anything else. Pick your two most repeated workflows and turn them into skills. That alone will change your daily Claude experience. Do not try to build 10 skills on day one. Start with

23:48 two. One for input, one for output. For example, one skill for research, one skill for review, or one skill for meeting summaries, one skill for email replies. The goal is not to collect skills. The goal is to remove repeated prompting from your workflow. Second, connect Gmail and Google Drive. The

24:08 OAuth takes two minutes and the capability jump is immediate. This is where Claude stops being a blank chat window and starts working with your actual context. Then open the browse catalog and add whatever else maps to your daily work. Slack, Notion, or anything from the MCP directory. Third,

24:26 web search and code execution are already built in. No installation, no toggle, they just work. Web search gives Claude access to current information whenever the task needs it. Code execution lets Claude actually run analysis instead of just explaining how analysis would work. Fourth, browse the

24:47 plugin directory once and install one or two that map to a workflow you already have. Enterprise search if you need cross-tool visibility. data if your work touches numbers. Remember the directory is dev-heavy right now. Pick precisely, don't install in bulk. Fifth, if you're ready to go beyond chat, install Claude

25:09 desktop and try co-work. Start with one Anthropic plugin that maps to your role. Marketing, sales, finance, data, whatever fits. You'll get a full stack of workflows and connectors in one install and you'll immediately see the difference between building skills here yourself and using ones designed for

25:27 your job. That's the moment the ecosystem stops being a bunch of features and starts feeling like a system. Sixth, if you work with code, openclaud.com/slashgems and spend an afternoon in the community MCP catalog. Pick one server that connects to a system you actually use. Do not start with a toy demo. Start with

25:49 something that already costs you time. A database you query manually, an internal API you dig through, a project tracker you check five times a day. Connect that one system through MCP and run a real workflow through it. That is how you find the value. And here's what I'd skip for now. Plugins that don't map to a

26:08 workflow you actually have today. The directory is growing fast, so check back, but don't install speculatively. Any MCP server you can't immediately map to a real use case. And anything you are installing only because it sounds futuristic. Your Claude environment gets noisy fast if you do. The full Claude

26:29 code guide is linked in the description and so is the co-work deep dive. Both are worth watching after this if you want to go further on either of those surfaces. I'll see you in the next one.