Claude Plugins vs. Skills: What’s the Difference and When to Use Each

Every team we have rolled Claude out to in the last six months has asked the same question in week one: "What’s the difference between a plugin and a Skill?" Sometimes followed by "and where does MCP fit in?" Anthropic’s documentation answers it correctly but in scattered pieces, and most third-party articles muddle the two. The result is teams who build a Skill when they should have installed a plugin, install a plugin when they should have written a Skill, and reach for MCP when neither is required.

This post is the explainer we wish existed when we started. One table, one decision tree, two worked examples. If you have read our Claude Skills guide and the Cowork business guide, this fills in the layer between them.

The one-line distinction

Anthropic’s own framing is the clearest:

Plugins are containers for distributing and organising (possibly several) Skills as part of a larger toolkit. Skills focus on providing specialised knowledge or procedures for tasks.

Translated for business users:

• A Skill is a recipe.
• A plugin is a cookbook that comes with a stocked pantry.

The recipe tells Claude how to do something. The cookbook is how you distribute a set of recipes to your whole team, often bundled with the ingredients (connectors) and tools (slash commands) needed to run them.

Which statement best describes the relationship between skills and plugins?

If you are here from an Anthropic Academy quiz, the correct answer is: a plugin bundles skills together so they can be installed in one package.

The other usual options are wrong for specific reasons:

• "They're two names for the same thing." No. A Skill is one piece of expertise; a plugin is the package that ships one or more Skills, often alongside connectors and slash commands.
• "Skills work in Cowork; plugins only work in Claude Code." No. Both are available in the Claude app and in Claude Code.
• "Plugins are built by Anthropic; skills are built by users." No. Anyone can build either. Anthropic publishes plugins, and you can build your own plugin to distribute your team's Skills.

The detailed comparison

What are Claude Skills?

Claude Skills are reusable expertise files Claude loads on demand. Each Skill is a folder containing a SKILL.md file with YAML frontmatter and markdown instructions, optionally accompanied by reference documents, templates, or helper scripts. When a user request matches the Skill’s description field, Claude pulls that Skill into context for the task at hand — and only for that task. Your context window stays clean, and one organisation can run fifty Skills without slowing anything down.

The practical use of Skills is to encode the kind of expertise a senior team member carries in their head. Brand voice, reporting formats, cold-outreach playbooks, contract review patterns, client-specific rules. Written once. Reused everywhere. Iterated weekly.

For a step-by-step walkthrough of building one, see our Claude Skills guide. For the highest-ROI Skills to start with, see our list of best Claude Skills for business.

What are Claude plugins?

Claude plugins are bundled packages of Skills, slash commands, MCP connectors, and agents — distributed as a single installable unit. Anthropic ships official plugins for HR, design, engineering, operations, finance, investment banking, equity research, private equity, and wealth management. Teams on Team or Enterprise plans can also build private plugins and distribute them through an organisation-wide marketplace with per-user provisioning and auto-install.

The Claude for Small Business launch in May 2026 is itself a plugin: 15 Skills plus 7 connectors (QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, Microsoft 365), packaged for one-click toggle install with no setup work required.

The mental shift: a plugin is infrastructure. You install it once and the whole team — or a slice of the team — gets the full bundle of expertise plus the tools to act on it.

What goes inside each

Inside a Skill

A Skill is a folder. The minimum is a SKILL.md file.

client-reporting/
  SKILL.md           ← instructions + YAML frontmatter (required)
  metrics-glossary.md← optional reference
  templates/
    monthly-report.md
    quarterly-report.md

The SKILL.md itself looks like this:

---
name: client-reporting
description: Generate monthly performance reports for ongoing client
  engagements. Use when the user asks for a monthly report, client
  update, performance summary, or wants to draft month-end deliverables.
---

# Client reporting

## Inputs
1. Client name
2. Reporting month
...

Claude loads the Skill only when the request matches the description. That single line is the entire trigger — the most common mistake is making it too vague ("creates reports" fails) or too narrow ("creates monthly reports for AcmeCorp" triggers for nothing else).

Inside a plugin

A plugin is a packaged bundle. The Anthropic HR plugin, for example, ships with:

• Skills: offer-letter, onboarding-plan, performance-review, compensation-analysis
• Slash commands: /draft-offer, /onboarding-plan, /perf-review
• Connector requirements: Google Workspace, BambooHR (or alternative HRIS), DocuSign
• Agents: sub-agents the parent plugin can delegate to for specific tasks

A plugin can be installed in one click from a marketplace. Behind the scenes Claude registers all the Skills, exposes the slash commands, and prompts the user to authorise the required connectors.

Where MCP fits in

Skills and plugins both describe how Claude should do something. MCP (Model Context Protocol) is how Claude connects to external tools to do it.

• MCP servers / connectors = real-time access to external systems. Read a HubSpot deal, send a Slack message, query Snowflake, post to Jira.
• Skills = the playbook for how to use that access well — what fields to read, what tone to write in, when to escalate.
• Plugins = a bundle containing Skills, slash commands, and the MCP connectors required to run them.

A cleaner way to think about it: MCP is for verbs, Skills are for nouns and adjectives, plugins are how you ship them together.

Decision tree: which one do I need?

Start at the top.

1. Is the customisation about what tool Claude can access?

• Yes: you need an MCP connector. Built-in ones cover 38+ tools. For anything else, see MCP server for non-developers.
• No, continue.

2. Is the customisation about how Claude should perform a specific task?

• Yes: you need a Skill. One folder, one SKILL.md, sharp description field. Continue.
• No, continue.

3. Do you need to distribute multiple related Skills + connectors + commands together, to a team or department?

• Yes: package them as a plugin and put it in your private marketplace.
• No: a Skill on its own is enough.

4. Does a public plugin already cover what you need (HR, finance, design, ops, IB, equity research, PE, wealth management)?

• Yes: install it. Customise with your own Skills on top.
• No: build your own plugin from a template.

Two worked examples

Example A: marketing team wants consistent blog drafting

The problem: five marketers, five different blog structures, no shared SEO discipline, hours wasted on inconsistent edits.

What to build: a blog-drafter Skill (one file, one team-wide upload), referencing a sandlabs-brand-voice Skill that every content Skill defers to. No plugin needed yet — a Skill is enough until you have ten of these and want to bundle.

Build time: thirty minutes. Time saved: four to six hours per long-form post.

Example B: investment-banking analyst onboarding

The problem: a boutique IB hires three analysts a quarter, each needing the same setup — comparable analysis, transaction document review, pitch material formatting, market research.

What to build (or install): the Anthropic Investment Banking plugin from the marketplace, customised with three internal Skills (firm pitch template, comp-set definitions, internal deal database). Distribute via the private marketplace with auto-install on day one of the analyst’s onboarding.

Build time for the customisation Skills: two days. Time saved: roughly two weeks of analyst ramp.

The most common mistakes

Building a plugin when you only need a Skill. Plugins are bundles. If you only have one piece of expertise to encode, a Skill is faster, cheaper, and more maintainable. Build the plugin when you have five or more related Skills that should travel together.

Writing a description field that does not trigger. "Writing assistant" is useless. "Drafts long-form blog posts in the Sandlabs founder voice with SEO keyword placement — use when the user asks to write, outline, or expand a blog post" triggers correctly. The description is the entire activation mechanism.

Forgetting that Skills and connectors are separate. A Skill that says "send the report to the client" cannot do anything without an MCP connector to Gmail or whatever email tool you use. Pair Skills with the connectors they need.

Trying to make Skills call other Skills. Skills cannot call each other programmatically. You can reference one Skill from another’s instructions ("follow the sandlabs-brand-voice Skill") but Claude decides whether to load that second Skill based on its own description match. Plan descriptions accordingly.

Skipping the org-wide deployment step. Skills uploaded to a single user are invisible to the rest of the team. For team rollouts, only Organisation Owners can provision org-wide Skills. If you build a great Skill and the team is not using it, check whether it was actually deployed at org level.

When to DIY versus bring in help

DIY works for one to five Skills, built-in plugins, and a single team. Most marketing or ops teams can be productive inside two weeks with the existing Anthropic plugin marketplace plus three or four custom Skills.

Bring in a professional when:

• You need a plugin library spanning multiple departments — twelve to twenty Skills bundled into three or four department-specific plugins.
• You want custom MCP connectors for a proprietary CRM, internal database, or undocumented API.
• You operate in a regulated industry and need version control, audit logs, and approval workflows over who can edit which Skills.
• You want Skills and plugins coordinated with custom AI agents that go beyond standard workflows.

At Sandlabs, our AI consulting engagements typically deliver a department-grade Skills library plus the MCPs to make it executable, in two to three weeks.

Tell us about your workflow →

The takeaway

The simplest mental model:

• MCP connectors give Claude access to do things.
• Skills teach Claude how to do them properly.
• Plugins package both into one-click installs for teams.

Most SMBs need a handful of MCP connectors, ten to fifteen custom Skills, and one or two plugins (either Anthropic’s public ones or a custom internal bundle). That stack covers ninety percent of business workflows. Build it deliberately and your team has Claude properly trained on your business inside thirty days — see our first 30 days rollout playbook for the sequencing.

Frequently Asked Questions

Which statement best describes the relationship between skills and plugins?

A plugin bundles skills together so they can be installed in one package. A Skill is a single piece of expertise - a SKILL.md file with instructions - while a plugin is the installable package that ships one or more Skills, often with MCP connectors and slash commands. They are not two names for the same thing, both work in the Claude app and Claude Code, and anyone can build either.

What is the difference between a Claude plugin and a skill?

A Skill teaches Claude how to do one task properly, such as drafting a report in your format. A plugin is a container that bundles several Skills with the connectors, slash commands and agents they need, so a whole team can install the set in one step. Build a Skill for a single workflow; build or install a plugin when five or more related Skills should travel together.

Do Claude plugins only work in Claude Code?

No. Plugins are available in the Claude app as well as in Claude Code. Organisations can distribute them through a private marketplace with auto-install, and Anthropic publishes official plugins for functions such as HR, finance, engineering and design.

Sitting a Claude certification? The Skill-versus-plugin distinction above is a live exam topic across the Associate and Architect tracks. Our Claude certification guide covers all four Anthropic exams, what each one tests, what they cost in AUD, and the honest answer on whether your team should bother.

Explore our AI & Claude consulting services →