Learn / Skills, connectors & plugins
Skills, connectors & plugins
These three words come up the moment you try to do real work with Claude, and most people cannot tell them apart. They are not the same thing, and knowing which is which is the difference between a chatbot and a tool that does your actual work. At the first AI Power Users meetup, 50 people who had never written a line of code set all three up in one evening. Here is the mental model they walked in with.
The 30-second version
Skills
A Skill is a reusable pack of instructions that teaches Claude to do one specific thing well, the same way every time. You set it up once, then call it whenever you need it. A Skill does not give Claude new data or new tools. It gives Claude a method. Think of it as a recipe Claude follows on command, so you stop re-explaining the same task. You are not limited to the ones other people share either: anyone can write their own, no code needed, because a skill is just a short text file. There is a guide to writing a good one further down this page.
From the meetupIn the room, a few people wrote a Skill on the spot that turned their messy meeting notes into the exact format they send clients. Built once, reused forever.
The skills worth installing first →Connectors
A Connector links Claude to a tool you already use, like Gmail, Google Calendar, Notion, or Xero. Once it is connected, Claude can read from that tool and act on it inside a normal conversation. This is the one most people install first, because it turns Claude from a clever chatbot into something that works with YOUR actual stuff. Under the hood this runs on MCP, the open standard for plugging tools into AI.
From the meetupMost people walked out of the meetup with at least one connector running. By the end, Claude was reading invoices, sorting calendars, and pulling notes from the tools they already lived in.
What MCP means, in plain English →Plugins
A Plugin is a bundle of the other two. It packages skills and connectors together so Claude can take on a whole area of work, not just one task. You install it once and Claude is ready across everything the bundle covers, instead of wiring up each tool and writing each skill yourself. Anthropic’s Claude for Small Business is a good example: one package of connectors and ready-to-run skills that spans finance, sales, marketing, HR, operations, and customer service, so Claude can chase invoices, close the month, and run a campaign.
From the meetupAt the meetup we walked through Anthropic’s Claude for Small Business and watched connectors like HubSpot, Intuit QuickBooks, and Canva go in. A whole toolkit landing in one step, rather than connecting each app by hand.
When to reach for which
| If you want | Reach for |
|---|---|
| Claude to do a repeatable task the same way every time | a Skill |
| Claude to see and act on your real data: email, calendar, books | a Connector |
| Claude ready across a whole area of work, with the skills and connectors already bundled | a Plugin |
Start here
The fastest win is one connector. Pick the tool you live in, usually Gmail or Google Calendar. In Claude, open the connectors menu, choose it, and approve access. Then ask Claude something it could only answer with that data, like “what meetings do I have tomorrow, and which ones have no agenda yet?” Once that lands, add the next one. That is exactly how the meetup ran: one connector each, live, before anyone touched skills or plugins.
Going further: writing a skill Claude will actually use
Connectors and plugins you install. Skills you create yourself, which sounds like the technical part but is the most approachable part of this whole page.
No need to write it by hand
Just describe what you want in plain English, like “make a skill that turns my meeting notes into a client update,” and Claude writes the properly formatted file for you: name, description, structure and all. Anthropic even ships a Skill Creator that walks you through it, and you can build one through a normal conversation.
How to use it Invoke it with /skill-creator and choose a mode: Create, Eval, Improve, or Benchmark. Then try a prompt like:
- “Create a skill that turns my meeting notes into a client update”
- “Create a new skill that reviews PRs for security issues”
- “Run evals on my code-review skill”
- “Improve my deploy skill based on these test cases”
- “Benchmark my skill across 10 runs and show variance”
The interactive workflow guides you through gathering requirements, creating test cases, and iteratively improving the skill until it works.
However you make it, these habits from Anthropic’s own guidance are what separate a skill Claude reaches for from one it ignores. They are also how you brief Claude well, and how you check what it hands back.
- Nail the description. A skill is just a text file that opens with a name and a one-line description. That description is the ONLY thing Claude reads to decide whether to use your skill at all. Say both what it does and when to use it, in the third person. Good“Turns meeting notes into a formatted client update email. Use when the user pastes notes and asks for a recap or follow-up.” Vague“Helps with emails.”
- Show an example, do not just describe one. Paste a real input and the exact output you want back. Claude copies a worked example far more reliably than it follows a paragraph of rules. This one habit does more than anything else on this list.
- One skill, one job. Name it after the job, like
client-update-from-notes, nothelperormy-stuff. If it is trying to do three things, it is three skills. - Keep it short, and assume Claude is smart. Only write down what Claude could not guess: your format, your rules, your house style. Skip the part explaining what an email is. Anthropic’s rule of thumb is to keep the file under 500 lines.
- Spell out steps only when the task is fragile. When many routes reach a good answer, give direction and trust Claude. When one exact sequence matters, write the steps out and tell it not to deviate.
- Test it on real work. Run it on actual tasks, watch where it gets things wrong, then fix the skill. It is never right on the first try.
Anatomy of a SKILL.md
-
name - A short kebab-case name: lowercase, hyphens, no spaces or capitals, up to 64 characters. Make it match the folder the skill lives in. It cannot contain "claude" or "anthropic".
-
description - The line that matters most, and the only part Claude reads to decide whether to use the skill. Say what it does AND when to use it, in the third person, using the phrases a user would actually type. Up to 1024 characters, but keep it tight.
-
metadataoptional - A block for extras like author, version, or a license such as MIT. Most skills do not need it.
Everything below the name and description is the body: a short overview, the instructions, and at least one worked example. A skill is really a folder, SKILL.md plus optional scripts/, references/, and assets/ that Claude opens only when a task needs them, so the name and description stay light. To share one, name the folder after the skill, add those files, and zip it. Anthropic’s full reference walks through every field: How to create custom skills.
---
name: client-update-from-notes
description: Turns rough meeting notes into a formatted client
update email. Use when the user pastes notes and asks for a
client recap, update, or follow-up.
---
# Client update from notes
Turn rough, messy notes into a short client update written in the
house style. Same shape, every time.
## Instructions
1. Sort every point in the notes into one of three buckets:
what is Done, what is Next, and what you Need from the client.
2. Write the subject as "Website update, [today's date]".
3. Greet the client by first name.
4. Write three sections in this order: Done, Next, Needs from
you. One short line each, plain language, no jargon.
5. Keep the whole email under 150 words and sign off as Keith.
## Format
- Subject: "Website update, [date]"
- Greeting: "Hi [first name],"
- Sections: Done / Next / Needs from you
- Sign-off: "Thanks, Keith"
## Example
Input:
design approved, waiting on logo files, dev starts Monday,
need their brand colours
Output:
Subject: Website update, 24 May
Hi Sarah,
Done: Homepage design approved.
Next: Development starts Monday.
Needs from you: Logo files and brand colours, so we can
lock the styling.
Thanks,
Keith
## When not to use
Skip this for internal updates or anything that is not going to
a client. Those have a different tone and structure.A name and description that say what and when, clear instructions, a real example, and a note on when NOT to use it. That is what a good skill looks like.
Want to set these up with help, not slides? I run hands-on workshops where you build and ship a real website with Claude Code in one day.