Claude Skills + MCP = AI Superpowers

Full video transcript

Welcome to the kickoff call for the Claude Mastery Sprint, the first of eight live workshops focused specifically on mastering Claude to make business and life better, easier, doing more of what you want to do and less of what you don’t. Claude has been a go-to tool for several years now, and it has continued to get better and better. It’s too good to ignore at this point. In an effort to help the entire community level up their use of AI and how to think about, architect, and build systems that support their goals in business, this month’s focus stays on Claude specifically. Over the next eight weeks there will be focused sessions on the foundation of Claude, how to architect the skills, recipes, and tools together inside of Claude to work with other tools and platforms, and most importantly how to approach building and creating with Claude, thinking through how far these tools have evolved compared to five years ago, three years ago, or even one year ago.

These platforms are evolving fast, with companies investing billions into making these tools better in a kind of AI arms race. Fortunately, using these tools doesn’t require being a billionaire. What’s needed instead is understanding what you want out of these tools and how to apply them to your business, how to serve customers and clients better, and how to do that without the overwhelm of chasing every new shiny object. This sprint is eight focus sessions on Claude specifically, with a challenge to approach it like an architect: orchestrating workflows and spending most of your time thinking through how these tools will actually work together.

The sessions will cover setting up a Claude account to be most effective. This may be redundant for people already working at a high level with Claude, but the goal is to build a solid foundation so nothing important is missed inside Claude’s traditional interface before moving into Claude Code, which will be another major focus later in the sprint. Claude Code will pair with the foundational material from this session.

One area of particular excitement is Claude Skills, essentially embedding recipes or outcome-focused prompts and prompt sequences that let Claude repeat a task with different context provided each time. All of this comes back to the human operator instructing these tools and orchestrating the overall system to make life better. Replays of all sessions will be posted inside the classroom in the Mastery series, and additional resources will be shared and built on throughout the month. For anyone new to Genai University, the “start here” foundation training is a good starting point, but the focus for this series is Claude Mastery specifically, continuing from where last year’s Gamma series left off.

The goal of the sprint is to meet weekly, focus on one tool, and collectively level up skills and abilities with something to show for it by actually doing the work and building with Claude, sharing learnings, challenges, and roadblocks live so they can be addressed weekly. Anyone watching replays can also ask questions inside the community.

Before diving into building, it’s worth spending extra time thinking through, writing out by hand, typing out, or speaking into a voice note what a workflow and a full end-to-end success picture actually looks like. Spending more time on the thinking and strategy before building saves more time in the long run. The best builders getting the best results from Claude and other AI tools spend more time on strategy and vision for the whole system before executing. If a tool isn’t ultimately serving the business, it’s worth questioning whether it’s worth doing in the first place. The recommendation is to start with whatever task is already the most consistent, top-of-mind priority, and use Claude to help produce a workflow and content that supports that end goal, rather than trying to cover Claude Code this early. If someone doesn’t know how to use Claude in general, there’s no reason to start with Claude Code right out of the gate.

This session, recorded on January 6, 2026, starts with the Claude interface itself. Claude is described as a superpowered AI language model with a large team continuously developing and releasing more capable models. A recent upgrade to the Max plan, roughly $100 a month, was made mainly for higher usage limits after repeatedly hitting rate limits on the $20 plan used through most of the past year. Most of what’s demonstrated in this session is achievable on the $20 or $25 a month plan, possibly with some usage limits. Opus 4.5 is currently Anthropic’s more advanced model, Sonnet 4.5 is described as fantastic, and Haiku 4.5 is used less. Models keep improving over time, similar to how ChatGPT and other tools have iterated through version numbers, and what’s built inside Claude only gets better as the underlying models improve.

The “plus” button in Claude’s interface lets users add files, connectors, and more, including uploading files or photos for Claude to read and summarize, understanding text and context in images, reading documents, and using a book as a knowledge reference to inform outputs. Content can also be added from Google Drive, GitHub, and Notion; once Notion is connected, Claude can update a Notion document directly from an instruction in chat without manual copy-paste, something that works far more reliably now than roughly a year ago. Claude also supports web search and custom user styles for controlling tone and verbiage.

Connectors are described as tools you give Claude so it can go and act on your behalf rather than requiring information to be manually brought in. Firecrawl is one example, a more advanced web research platform, connected directly for research and testing over the past few weeks. The Zapier MCP server was highlighted specifically: setting up a single Zapier MCP connection allows access to 8,000+ apps that Zapier integrates with, rather than configuring individual MCP servers for every tool. Setting up a new Zapier MCP server involves creating a server, selecting a client, and then adding tools such as a CRM lead connector, Notion, Slack, or HubSpot; once added, each of those tools’ actions become available without further setup on the Claude side. This was described as a shortcut for accessing several tools without setting up every individual MCP connector.

Manage Connectors is found under account settings in the bottom left corner, alongside profile settings, work function description, privacy controls (including a toggle for whether chats are used to help train Anthropic’s models, similar to a ChatGPT data-training toggle), and options to export chats or manage memory preferences. A question was raised about connecting apps that only have free accounts; the answer is that usage limits vary by tool and depend on that platform’s own terms of use, since some tools require a paid plan to use their MCP integration at all.

On the tradeoff between connecting a tool directly to Claude versus adding it as a tool inside Zapier MCP: using Zapier for something like Firecrawl burned credits quickly, so Firecrawl was connected directly instead, reducing usage considerably. Direct native connections are recommended when available; Zapier is useful to fill gaps for tools that don’t have their own connector.

Moving into capabilities: an artifact is generated content, like code snippets, text documents, or website designs, that appears in a dedicated window alongside the conversation. Artifacts can be made private or public, shared, and embedded elsewhere. There are also AI-powered artifacts for building apps and prototypes, toggled on in settings, along with code execution and file settings, and a linked security risk document covering what domains a sandbox can access.

Two more settings were covered before moving into skills. Memory, if toggled on, lets Claude search for relevant information and details from past chats across the different threads in the sidebar, which is optional depending on individual preference. A participant asked whether skills set up in the standard Claude interface are also referenced by Claude Code, since Claude Code has its own slash commands sometimes referred to as skills; the answer given was that this isn’t fully confirmed yet. The working theory is that skills built in a Claude account could or should be accessible by Claude Code, and that skills may need to live inside a GitHub repository or locally so Claude Code can access them, but this is something still being tested going into upcoming sessions.

A breakthrough over the past week and a half was in understanding how skills actually work. The comparison was made to building recipes with Jasper years ago, when a well-built recipe still required copying and pasting every prompt in sequence each time. With Claude, skills solve that: build a good instruction once, and Claude can execute it repeatedly. A set of recipes has been created and will be shared with the community, giving members in sales, marketing, or operations a starting point they can build on rather than starting from zero. The challenge is not to get overwhelmed by the number of available recipes, but to focus on the ones matching a specific role or need.

A live demonstration followed using the “Market Sauce Blueprint” skill: given a website link, Claude used its Firecrawl connector to research the site, referenced the Market Sauce Blueprint skill to understand the framework, and produced a multi-page blueprint document, mirroring a process that used to take 30 minutes to a few hours done manually. This blueprint asset was described as foundational, something a consultant or agency could use as a first deliverable for a client, later feeding into follow-on skills like generating an email campaign or sales copy that use the blueprint as context. The skill file itself was shown structured as a skill.md with a description, contents, required inputs, instructions for what to do if information is missing, example outputs, and the recipe instructions.

Several participant questions were addressed live. On whether a Claude skill can be built from an existing framework or a book: yes, a PDF or existing document can be uploaded and turned into a skill by asking Claude to convert it into that structure. On multi-repository Claude Code tasks: current experience is that Claude Code references one GitHub repository at a time, and whether it can span multiple repositories for a single feature request wasn’t confirmed, noted as something to test in future Claude Code-focused sessions. On structuring multiple skills for a multi-category service (e.g. detoxing home, body, mind, and nutrition): the suggested approach is to bucket each area as its own skill, then have a master agent call on the relevant skills as needed, or alternatively give the master agent access to all the skills at once and let it sequence them, testing both approaches to see which performs better. On exporting Claude outputs into a CRM like GoHighLevel: this requires that CRM’s connector (for example, a lead connector MCP) to be added, after which Claude can be instructed to update a specific contact’s record with new information, similar to a demonstrated example where a 90-day plan was pushed directly into a Notion document without manual copy-paste, something noted as far more reliable than roughly 12 months ago. On team account settings for skills and projects: a team owner can control whether a project or skill built by one team member stays private to their account or is visible to the whole team.

The demonstration continued with Claude projects, shown as a way to keep information scoped to a specific client or venture. A new project was created, a blueprint document was uploaded to it, and the project was described as functioning like “a mini brain for Claude,” where uploaded content and custom instructions stay contextually tied to that project rather than the general workspace. From inside a project, a skill can be executed directly using the uploaded material as reference context.

Closing out the session, several resources were shared: two Claude quick-start guides (one PDF, one Google document) covering general Claude and Claude Code setup and foundational learning materials; a skills readme document outlining available skills across sales, marketing, and operations as a starting point for building custom skills; and a “Market Sauce Recipe Cookbook” containing a basic business-brief template (who you are, who you serve, unfair advantage, tone of voice) plus a series of stacked recipes such as researching a prospect, writing a cold email, handling objections, and building a follow-up sequence, each usable individually or combined. The framing offered was to think about recurring, well-defined processes, like writing SOPs repeatedly, as candidates for turning into a reusable skill, buying back the time spent on that task every time it recurs.

A final live demonstration showed building a new skill using Claude’s own skill-creator tool: uploading an existing 28-page consolidated document combining several prior recipes (buyer psychology, competitor intelligence research, and opportunity identification into one report), and letting Claude’s skill-creator ask clarifying questions about what to build, then generate and “bake” the new skill. The finished skill was saved directly to the account’s skill library and confirmed to be immediately runnable using a trigger phrase.

The call wrapped with a preview of the next session, focused on recipe development, building out more skills, and open Q&A, with Claude Code sessions coming in subsequent weeks. The reasoning given for saving Claude Code until later is that it makes little sense to start there before understanding the basic components of connectors and skills covered in this kickoff session.