AI Guides › Step-by-step guides
By Nigel Guy · 6 min read
Most people meet Claude's extras as a pile: plugins, skills, connectors, MCP servers, all in one "install these 24 things" list. They install a dozen in an afternoon, use two, and leave the rest quietly adding context cost and, in some cases, running code on their machine. The pile feels productive because installing is fast. The trouble starts when you can't say which layer a thing belongs to, or what it is allowed to touch.
The rule: pick one job, install one thing from the right layer, use it for a week, and only then add the next.
A note on scope. A "24 things" list was the starting point for this guide, but I have only included items I could confirm in Anthropic's own docs and repositories on 2026-10-04. Third-party tools in those lists change fast and I could not verify them, so they are left out. The counts below are smaller than 8-8-8 on purpose.
First, know the three layers.
| Layer | What it is | Where you manage it |
|---|---|---|
| Skill | A folder of instructions (a SKILL.md plus optional scripts) Claude loads when the task fits |
claude.ai: Customize > Skills. Claude Code: via plugins or your skills folder |
| MCP connection (connector) | A link to an outside service so Claude can read data or take actions there | claude.ai: "+" in chat > Connectors > Manage connectors. Claude Code: claude mcp add and /mcp |
| Plugin | A bundle that can contain skills, slash commands, agents, hooks and MCP servers | Claude Code: /plugin. Cowork and claude.ai have their own plugin pages |
What you need:
Write one sentence: "I want Claude to ___." Then choose:
Skills Anthropic publishes in its anthropics/skills repository include docx, pdf, pptx, xlsx, doc-coauthoring, frontend-design, internal-comms, brand-guidelines, theme-factory, canvas-design, webapp-testing, web-artifacts-builder, mcp-builder, skill-creator and slack-gif-creator. Anthropic's document skills (Excel, Word, PowerPoint, PDF) are built in and Claude applies them automatically when relevant.
On claude.ai:
In Claude Code, add the repository as a marketplace, then install a set:
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
Starter picks from the official marketplace (all folder names confirmed in anthropics/claude-plugins-official): commit-commands, code-review, feature-dev, frontend-design, security-guidance, pr-review-toolkit, claude-md-management, hookify, skill-creator. The official marketplace also holds language-server plugins such as pyright-lsp and typescript-lsp.
For non-coding work, Anthropic's knowledge-work-plugins include productivity, sales, customer-support, product-management, marketing, legal, finance, data and enterprise-search.
/plugin install commit-commands@claude-plugins-official. In a session this opens the plugin's details rather than installing at once./reload-plugins.For knowledge-work plugins, add the marketplace first:
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install sales@knowledge-work-plugins
In Cowork, install from claude.com/plugins instead.
The easiest route is the directory: in claude.ai click + in chat, Connectors, Manage connectors, pick one and sign in. Examples named in Anthropic's help article are Linear, Slack and Google Drive. Claude inherits your permissions in that service, so it cannot reach what you cannot.
In Claude Code, the docs show these commands:
claude mcp add --transport http notion https://mcp.notion.com/mcp
claude mcp list
/mcp
Use /mcp to sign in when a server shows "Needs authentication". If you are signed in with a claude.ai account, connectors you set up at claude.ai/customize/connectors appear in Claude Code automatically. GitHub, Stripe and HubSpot also appear as examples in the docs; check each vendor's current URL and auth method before copying anything.
Before anything beyond Anthropic's own, paste this into Claude.
You are a cautious reviewer helping a non-expert decide whether to install a Claude add-on.
Add-on: [NAME_AND_SOURCE_URL]
Layer: [SKILL, PLUGIN OR MCP_CONNECTOR]
My job to be done: [ONE_SENTENCE]
My plan and tools: [PLAN, CLAUDE_CODE_OR_CLAUDE_AI]
Task: from the files I paste below, say what it can read, write, run or send, and whether that matches my job. List any hooks, scripts, bin/ files, or servers and what each does.
Files pasted: [SKILL.MD / hooks.json / .mcp.json CONTENTS]
Output: a table (component, what it does, risk: low/medium/high), then a one-line verdict: install, install with changes, or skip.
Rules: if I have not pasted a file you need, ask for it instead of guessing. Do not assume the add-on is safe because it is popular. Before answering, check that every claim comes from the pasted text.
Fill in the name, layer, job and the pasted files.
/ and look for /<plugin>:<skill>, or run claude plugin list in your shell./mcp and look for "Connected"; then ask Claude a read-only question, such as listing your last five items.claude plugin uninstall, claude mcp remove <name>, or switch the skill off.