Claude Skills vs Plugins vs Projects: Which One Your Business Needs (2026)
Claude skills vs plugins vs projects in plain English: what each one holds, which plan includes it, and a one-line rule for picking the right one per job.
Open Claude's Customize menu and you'll see Skills, Plugins and Projects sitting side by side, with Connectors underneath and Memory somewhere in Settings. The names sound like five ways to do one thing. They aren't, and picking the wrong one is how a business ends up with a Project per client, each holding a copy of the same instructions that drifted apart in March.
Claude skills vs plugins vs projects comes down to what each one holds. A Project holds context: your files, your rules, the chats about them. A skill holds a procedure: how you do one job, written once. A plugin holds a box of skills, plus the connectors they need, sold as a bundle. Get those three sentences straight and the Customize menu stops being a puzzle.
Below: what each one is, which plan includes it (read from Anthropic's help pages on November 2, 2026), where the others fit, and the one-line rule we give clients, with a worked example.
Claude skills vs plugins vs projects, item by item
A Project is a self-contained workspace with its own chat history, its own instructions and its own knowledge files. Everything in it is loaded whenever you start a chat inside it. Free accounts can create five; paid plans list no cap and add retrieval for knowledge bases that outgrow the context window. Our lesson on how to use Claude Projects names the five worth building first.
A skill is a folder with one file in it, SKILL.md, that describes when to use it and the steps to follow. Claude reads the description, and when a request matches ("categorize these transactions"), it loads the full file and works from it. Skills are on every plan, Free included; you switch on code execution under Settings > Capabilities and upload the ZIP under Customize > Skills. They also run in Claude Code and the Excel, Word, PowerPoint and Outlook add-ins, and the same file loads in ChatGPT, Gemini and Copilot.
A plugin is a package. Anthropic's help page says each one bundles skills, connectors and sub-agents, and some add commands you can type with a slash. You install one from a marketplace under Customize > Plugins > Discover, and plugins are available on paid plans only: Pro, Max, Ultra, Team and Enterprise. The best-known one for owners is the Claude for Small Business plugin, which packed 44 skills and 35 app connections when we read its repo in October.
So a Project is where you work, a skill is how a job gets done, and a plugin is a shortcut to installing several skills and their connections at once.
Projects: the filing cabinet
Say you run a two-person landscaping company. Your Quotes Project holds the rate card, last season's three best quotes as examples, and an instruction that says exclusions come before payment terms. Every chat you open inside it starts knowing all of that. Your Customer Replies Project holds the refund policy and the tone you want, and nothing about rates.
That's the job of a Project: background that stays true for one kind of work, kept where you can read it and change it. Anthropic's help page describes it as knowledge that's always loaded when you start chats within it. The limit is the flip side of the same design. What you put in a Project is only there inside that Project, so a rule you need in three places gets copied three times, and copies drift.
Two things worth knowing. Project sharing is a Team and Enterprise feature, so on Pro your office manager can't open your Quotes Project. And Anthropic is testing a new version of Projects, starting in Claude Code, where a project is one long conversation with parallel threads. Existing Projects carry over, but the shape may change under you this year.
Skills: the procedure card
A skill is the thing that stops the copying. Our bookkeeping copilot is about 950 words: it asks for your chart of accounts once, refuses to invent a number, flags statements that end mid-month, and closes every answer by telling you to have your accountant check anything important. Install it once and it loads in the Books Project, in a plain chat, and in Cowork, whenever the request looks like bookkeeping.
Anthropic's own description is the useful one: Projects give static background knowledge, skills give specialized procedures that activate dynamically when needed. The company's March 2026 explainer on how skills compare to prompts, Projects and MCP adds the rule of thumb. If you keep retyping the same prompt, make it a skill. If you keep copying the same instructions into several Projects, make it a skill.
A skill can't reach your inbox or your accounting software by itself. It works from whatever you paste or upload, which is also why you can read every line of it before it touches a customer's data. The 55 in the Vexlo library are built that way on purpose.
Plugins: the box
A plugin saves you the installing. One click under Customize > Plugins > Discover and you have a set of skills, the connectors they expect, and in Cowork the sub-agents and hooks that let a job run in several steps with an approval at each one. Hooks and sub-agents run in Cowork and Claude Code, not in ordinary chat, so a plugin does less in the web chat than its listing suggests.
The catch for a small business is the plan. Plugins need a paid seat, and the plugin most owners will meet, Anthropic's own Claude for Small Business, is built around connectors to QuickBooks, HubSpot, Google Workspace and Microsoft 365. If your books live in QuickBooks and you pay for Claude, it's worth the two-minute install. If you're on the free plan, or your books are in Wave or a spreadsheet, most of the box stays shut, and single skills cover the same jobs.
One more thing the help page says plainly: only install plugins from sources you trust. A plugin can carry a local connector that runs with the same permissions as any other program on your computer. Read the marketplace name before you click Add.
Connectors, memory and artifacts, so you stop confusing them
Connectors (built on MCP, the standard Anthropic uses for them) give Claude access to a system: Gmail, Google Drive, Outlook, QuickBooks. A connector is the door; a skill is what Claude does once it's through it. Anthropic's explainer says the two are usually used together, and that matches what we see. Connect Gmail, Drive and Outlook safely covers the settings. Connectors are on the free plan.
Memory saves small facts from your chats and brings them into later ones, on every plan. It suits preferences and habits. A price, a policy or a legal commitment belongs in a Project file where you can read it, which the memory lesson lays out with a four-row table.
Artifacts are outputs: the document, slide deck or spreadsheet Claude produces. People sometimes save instructions in one and wonder why Claude forgets them. An artifact is the thing you made, not a place Claude reads from.
The one-line rule, and a worked example
Facts about your business go in a Project. Procedures go in a skill. Access to a system is a connector. A plugin is a bundle of the last two, bought with a paid seat.
Back to the landscaping company. The owner builds two Projects: Quotes (rate card, three example quotes, the exclusions-first rule) and Customer Replies (refund policy, tone). She installs one skill, the job estimate builder, which turns site notes into a numbered scope with allowances and a validity date; it runs inside Quotes and, when a customer emails asking for a price change, inside Customer Replies too. She connects Gmail so Claude can draft replies in the thread. That's two Projects, one skill, one connector, on the free plan, with nothing bought.
In spring she hires an estimator and moves the books to QuickBooks. Now a paid seat makes sense, the Team plan math is worth reading, and the Claude for Small Business plugin earns its install for the QuickBooks side. The Projects and the skill stay as they were.
If your business has prompts scattered across six note files, the prompt library builder turns them into one maintained set first, and the ones you use weekly are your skill candidates.
Where this breaks
The rule assumes you'll keep the layers apart, and the pull is always toward the Project. It's the visible one, so instructions pile up in it until each Project is a 3,000-word document nobody rereads. Once a quarter, read your Project instructions and move anything that says "how to" into a skill.
Skills don't manage themselves either. A skill written in July still says July's prices if you put prices in it, which is why prices belong in the Project file and not the procedure. And a skill from a stranger is a set of instructions your assistant will follow; run it through Skill Check before you install it.
Plugins change fastest. The plan rules, the default marketplace and the sub-agent behaviour were all read on November 2, 2026, and Anthropic has moved features between plans twice this year. Check the plugin's listing on the day you install it.
If you're setting Claude up from scratch, the free Claude for Small Business course builds the Projects first, then adds skills and connectors in the order above. The ranked list of ten skills is where most owners pick their first one.
Common questions
- What is the difference between a Claude skill and a Project?
- A Project is a workspace: its own instructions, its own files and its own chat history, always loaded when you work inside it. A skill is a procedure in a file that Claude loads on its own when a request matches, in any chat and in any Project. Anthropic's help pages put it as background knowledge that is always there against specialized procedures that activate when needed. Keep your price list in a Project; keep how you write a quote in a skill.
- Are Claude skills and plugins the same thing?
- No. A skill is one folder with a SKILL.md file inside. A plugin is a package that bundles several skills with connectors, commands and sub-agents, installed from a marketplace under Customize > Plugins. Skills come with every plan, Free included. Plugins need a paid plan (Pro, Max, Ultra, Team or Enterprise) as of November 2026.
- Do Claude skills work on the free plan?
- Yes. Anthropic's help centre lists skills on Free, Pro, Max, Ultra, Team and Enterprise. You turn on code execution under Settings > Capabilities, then upload a skill as a ZIP under Customize > Skills. The same file also loads in ChatGPT, Gemini and Copilot.
- Can I use a skill inside a Project?
- Yes, and that is the normal setup. A skill is not tied to one Project. If you find yourself pasting the same instructions into three Projects, move them into a skill once and every Project picks it up when the job comes up.
- How many Projects can I have on the free Claude plan?
- Five, as of November 2026. Paid plans list no cap and add retrieval for large knowledge bases. Five is enough for one small business: customer replies, quotes, marketing, the books and a spare.
- Claude skills vs plugins vs MCP connectors: which do I need?
- A connector (built on MCP) gives Claude access to a system, such as Gmail, Google Drive or QuickBooks. A skill tells Claude the procedure for working with what it finds there. A plugin ships the two together. Most owners need a couple of connectors and a handful of skills before they need a plugin.
More from the blog
Want this customized and automated for your business?
We take the tools in this toolbox and wire them into your business: your data, your brand voice, running on autopilot.
Talk to Vexlo