Connectors. Skills. Plugins. MCPs. Agents. Cowork. Artifacts. Memory. The plain-English dictionary of every AI term you'll actually run into this year, plus how each one fits together.
Most people opening Claude in 2026 hit the same wall. Connectors, Skills, Plugins, MCPs, Cowork, Artifacts, Memory. Each of these is a real, separate feature inside Claude. They sound the same. They are not.
This guide does two things. First, it nails down the three most-confused features in three plain sentences. Second, it gives you a one-page dictionary of every other AI term you'll bump into this year. Save it. Send it to the friend who keeps asking what an MCP is.
A connector is the cable. It plugs Claude into the apps you already use. Gmail, Google Calendar, Drive, Slack, Notion, Shopify. When a connector is on, Claude can read what's in that app and, in many cases, take action inside it. No connectors, no real-world reach. Claude is just a chatbot in a box.
A skill is the manual. A skill teaches Claude how you want a specific task done. Once. You write the instructions, give it a name, and from that point on Claude follows that exact workflow every time you call it. "Format this PDF the way I always format PDFs." "Write this email in my brand voice." "Pull weekly highlights from my calendar the way I like them." The skill remembers how. You stop re-explaining.
A plugin is the package. A plugin bundles a stack of skills plus the connectors they need into one install. Instead of building a marketing workflow from scratch, you install a marketing plugin and the campaign-builder skill, the social-scheduler skill, the brand-voice skill, and the connectors to Klaviyo and Meta all show up wired together.
Paste this into Claude any time you hear an AI term you don't recognize. Claude will explain it in plain English, tell you when you'd actually use it, and flag whether it's something a non-technical person should care about or safely ignore.
Copy this prompt
You are my AI terms translator. Whenever I drop a term into this chat, explain it back to me in this exact format: 1. PLAIN ENGLISH DEFINITION. One sentence. No jargon. No "leverages." No "synergies." Talk like you would to a smart adult who doesn't work in tech. 2. THE ANALOGY. One concrete analogy from real life that makes the concept stick. (Example for MCP: "It's a USB port that lets Claude plug into a tool.") 3. WHEN I'D ACTUALLY USE IT. One specific scenario where a non-technical person would touch this term in real life. If it's something only developers care about, say so. 4. WHAT IT IS NOT. The one thing this term most often gets confused with, and the actual difference. (Example for Skill vs Plugin: "A Skill is one task, a Plugin is a bundle of skills.") 5. SHOULD I CARE? Either: "Yes, learn this now." Or: "Yes, but only when you're ready for X." Or: "No, this is developer-only — safely ignore." Be ruthlessly plain. If the term is a marketing word with no real meaning (like "synergistic AI") say so. The goal is for me to leave the chat understanding the term well enough to teach it to someone else.
The one-line cheat sheet
Connector = how Claude reaches your tools. Skill = how Claude does your task. Plugin = a ready-made workflow that bundles both. If you only remember three sentences about Claude this year, those are the three.
Connector. The cable from Claude into one of your apps. Gmail, Drive, Slack. Once it's on, Claude can read and (sometimes) act inside that app.
Skill. A reusable workflow you teach Claude once. From then on, Claude follows it the same way every time. Your brand voice, your meeting-prep format, your weekly-report shape.
Plugin. A bundle of skills and connectors installed together. Think "install a marketing department" instead of building each piece one by one.
Project. A long-running workspace inside Claude with its own files, instructions, and memory. Different from a regular chat because the context sticks around between conversations.
Memory. Persistent context Claude builds up across chats about who you are and how you work. Helpful when accurate, annoying when stale. (See the memory audit guide.)
Artifact. A standalone piece of output Claude makes for you. A doc, a slide, a webpage, an interactive dashboard. Lives inside the chat as its own object you can edit, copy, and share.
Cowork. Claude's collaborative workspace. The version where multiple people on your team can work alongside Claude on shared projects and skills.
MCP (Model Context Protocol). A standard way for Claude to connect to third-party tools that don't have a native connector yet. Think of it as a USB port for Claude. Apps like Shopify, Klaviyo, and Zapier ship MCP servers Claude can plug into.
Agent. An AI that takes multiple actions on your behalf without you having to prompt each step. "Book me a flight" instead of "search for flights, compare prices, fill in my passport details, click pay."
Connector vs MCP. A connector is a native, one-click integration Anthropic built. An MCP is the more open standard for anything that doesn't have a connector yet. Functionally similar to a user, technically different under the hood.
Prompt. What you type into Claude. The instructions, the context, the question. Better prompts = better answers. Most AI quality problems trace back to a thin prompt.
System prompt. A prompt that sits underneath every chat in a Project or Skill, setting the rules and tone Claude follows by default. You write it once, it shapes every reply.
Hallucination. When AI confidently makes up a fact that sounds real but isn't. The fix is asking Claude to say "I don't know" instead of guessing. (See the 3 ways AI lies guide.)
Sycophancy. When AI agrees with you when it shouldn't. The fix is telling Claude to argue against you, not for you.
Bias. When AI gives you skewed answers because its training data leaned one direction. The fix is asking Claude to make the strongest case for the opposite view.
RAG (Retrieval-Augmented Generation). A way to give AI access to your specific docs (instead of only its training data) so it can answer questions about your business, your files, your knowledge. Developer-flavored term, but useful to recognize.
Token. The unit of text AI reads and writes. Roughly half a word. Matters when you're hitting context limits or paying per use.
Context window. How much text Claude can see at one time. Bigger window = Claude can read longer docs without forgetting the start. Claude's window is large; you usually don't hit it on normal work.
Fine-tuning. Training a model on your specific data so it gets better at your specific use case. Mostly a developer move. Most non-technical people can get 95% of the benefit with a good skill or a good Project instead.
Scheduled task. A workflow you set to run on a recurring schedule. Every Monday morning, every Friday afternoon. The thing turns Claude from "chatbot you message" into "assistant who shows up on time."
The rule for new terms
If a term doesn't fit cleanly into one of the categories above (cable / manual / package / workspace / output / standard / action / quality issue), the term is probably marketing language, not a real feature. Treat it with suspicion until someone shows you what it actually does in real life.
The Only AI Masterclass You Need
If this guide helped, but you’re looking to go deeper, I got you!! My 30-Day Challenge takes you from saving AI tips you never use to actually building with AI, step-by-step.
I show you exactly how I automated two e-commerce brands, my social media, and most of my personal life, then hand you the agents, workflows & systems to do the same. I’m teaching you every single thing I know with one lesson and one build a day.
Join the AI Masterclass →© 2026 Mariah Brunner. All rights reserved.