Connectors. Skills. Plugins. MCP. Agents. Cowork. Artifacts. Memory. The plain-English dictionary of every AI term you’ll actually run into this year, and how they fit together.
Most people opening Claude in 2026 hit the same wall. Connectors, skills, plugins, MCP. They sound alike, and they’re all different things.
So here are the three most-confused ones in plain English, then a one-page dictionary for everything else. Save it, and send it to the friend who keeps asking what an MCP is.
01
A connector is the cable
It plugs Claude into apps you already use, like Gmail, Google Calendar, Drive, Slack or Notion. With a connector on, Claude can read what’s in that app and, for many apps, take actions there too. Without connectors, Claude only knows what you paste in.
02
A skill is the manual
A skill teaches Claude how you want one specific task done. You write the steps once, and Claude follows them whenever the task comes up. "Write this in my brand voice." "Format my weekly report the way I like it." You stop re-explaining.
03
A plugin is the package
A plugin bundles skills, connectors and other pieces for a role or team into one install. Add a sales or marketing plugin and the workflows show up already wired together. Plugins are on paid plans. (Claude Help Center)
Where To Find Them
In Claude, all three live in the left sidebar under Customize. Anthropic is renaming that menu Plugins and merging the Skills, Connectors and Plugins tabs into one page, so look for either name.
Project. A workspace in Claude with its own files and instructions. Every chat inside it starts with that context, so you don’t re-upload the same docs.
Memory. What Claude remembers about you across chats, like your job and how you like answers. Helpful when it’s accurate, annoying when it’s stale. In Claude it lives under Settings, then Your Claude. (See the memory audit guide.)
Artifact. A standalone thing Claude makes for you, like a doc, a webpage or an interactive dashboard. It opens next to the chat so you can edit, copy or share it.
Cowork. Claude’s mode for bigger, multi-step work on your files and apps, the kind that runs while you do something else. Anthropic is now folding it into regular Claude chat, starting with Pro, Max and Ultra, so on those accounts you can just ask for it in any conversation. (Anthropic)
MCP (Model Context Protocol). The open standard that lets AI connect to outside tools. Think of it as a universal plug. Claude’s connectors run on it, and companies like Shopify, Klaviyo and Zapier offer MCP servers you can plug in. (What is MCP)
Connector vs MCP. A connector is the ready-made, click-to-add version you pick from Claude’s directory. MCP is the standard underneath it, and you can add any MCP server yourself as a custom connector by pasting its URL.
Agent. An AI that takes several steps on its own to finish a goal. "Find me three flights under $400 and compare them" instead of you doing each search yourself.
Scheduled task. A job you set to run on repeat, like every Monday at 7am. It turns Claude from a chatbot you message into an assistant that shows up on time. Paid plans only.
Prompt. What you type into the AI: the instruction, the context and the question. Most bad answers trace back to a thin prompt.
System prompt. Instructions that sit underneath every chat and set the rules and tone. In a Claude Project, that’s the project instructions box.
Hallucination. When AI confidently makes up something that sounds real. Ask it to cite sources and 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. Tell it to argue against your idea and find the holes.
Bias. Skewed answers that come from skewed training data. Ask for the strongest case for the opposite view.
RAG. Short for retrieval-augmented generation. The AI looks things up in your documents before answering, instead of relying only on what it learned in training. Claude Projects switch to it on their own once your files get too big to fit in one go.
Token. The small chunks of text AI reads and writes. For Claude, one token is about 3 to 4 letters of English. Usage limits and API pricing are counted in tokens.
Context window. How much text the AI can hold in mind at once, counted in tokens. When a chat gets very long, early details can drop out, so start fresh chats for new topics.
Fine-tuning. Retraining a model on your own data. A developer move. For most people, a good skill or Project gets you most of the way.
New AI words show up every week. Paste this into Claude or ChatGPT once, then drop in any term you don’t recognize.
The One-Line Cheat Sheet
Connector = how Claude reaches your tools. Skill = how Claude does your task. Plugin = a ready-made package of both. If a new term doesn’t fit anywhere on this page, it’s probably marketing until someone shows you what it does.