Claude cowork skills and plug-ins

If this is your first pass, follow this exact order in a fresh project folder. You can stop after each step and still be in a safe place.

1. Set up the core tools (5 minutes): install claude, git, and a terminal you can already type in.
2. Create your first instruction file (5 minutes):

   cd your-project
   cat > CLAUDE.md <<'EOF'
   # CLAUDE.md
   ## Boundaries
   - Use short commands and diff-based edits.
   - Commit before risky changes.
   - Ask for explicit confirmation before write-side integrations.
   EOF

3. Run one baseline workflow (8 minutes): launch claude in the repo and ask for a tiny code-review task on one file.
4. Add exactly one integration (10 minutes): choose either one official marketplace plugin or one connector, not both, and keep it read-only first.
5. Validate and lock scope (2 minutes): run a quick git diff, confirm files/paths, then pause before adding MCP or write actions.

• Start with core workflow: CLAUDE.md + Claude Code + Git.
• Use official sources first: official docs/marketplaces before community options.
• Use least privilege: start integrations with read-only behavior, escalate only if needed.
• Keep secrets out of instruction files: do not paste tokens into CLAUDE.md.
• Separate concepts: connectors and MCP solve different needs; don’t assume they are the same.

Use one path first, then add complexity only if needed.

This guide is for beginners through advanced operators who want one playbook for setup, command habits, reusable Markdown instructions, MCP integration, and connector decisions across macOS, Linux, and Windows.

• Common free-first path: CLAUDE.md + Claude Code CLI + Git + one official connector or official marketplace plugin.
• Optional paid upgrades: Claude plans, SaaS connectors, and managed enterprise controls.
• Safety posture: treat every connector/MCP server as a trust decision, not a default.

Use this quick chooser when deciding between instruction files, plugins, connectors, and MCP.

Where each thing actually lives (surface map)

Scope cheatsheet (local / project / user-global / managed)

Use this as the fastest practical baseline. Start with CLI + Git, then add connectors/MCP only when needed.

# 1) Install Claude Code (official installer)
curl -fsSL https://claude.ai/install.sh | bash

# 2) Optional: install via Homebrew stable channel
brew install --cask claude-code

# 3) Launch in your project
cd your-project
claude

# 4) Verify CLI health
claude --version

# 1) Install Claude Code (official installer)
curl -fsSL https://claude.ai/install.sh | bash

# 2) Install core prerequisites
sudo apt update
sudo apt install -y git curl jq ripgrep fd-find

# 3) Launch in your project
cd your-project
claude

:: 1) Install Claude Code from CMD (official CMD script)
curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

:: 2) Optional: install via WinGet
winget install Anthropic.ClaudeCode

:: 3) Install Git for Windows (recommended for Bash tool)
winget install Git.Git

:: 4) Launch Claude
cd your-project
claude

• Adding plugins, connectors, and MCP together on day one usually creates too much complexity to debug.
• Putting secrets in CLAUDE.md leaks them into prompts and history.
• Granting write access before you validate a read-only flow increases recovery risk.
• Installing random community plugins/connectors before official options can skip useful defaults.
• Assuming connectors, plugins, and MCP are interchangeable usually causes integration confusion later.

Plugins are packaged extensions for Claude Code (skills, agents, hooks, and MCP server integrations) distributed through marketplaces.

Common categories and examples

• Code intelligence (explicitly documented): LSP-based plugins that enable jump-to-definition, references, and type diagnostics.
• Git workflow helpers: commit/PR workflow skills and guardrails (example: commit-commands in demo marketplace).
• System or integration utilities: plugins that package repeatable helper skills, hooks, or MCP wiring for known tools.

Catalog categories evolve. Re-check the Discover tab for current labels and install scopes before standardizing team workflows.

Verified install commands

# Open plugin manager
/plugin

# Install from official Anthropic marketplace (already available by default)
/plugin install github@claude-plugins-official

# If official marketplace looks stale/missing
/plugin marketplace update claude-plugins-official
/plugin marketplace add anthropics/claude-plugins-official

# Add Anthropic demo marketplace (manual)
/plugin marketplace add anthropics/claude-code

# Install demo plugin example
/plugin install commit-commands@anthropics-claude-code

# Activate new plugin versions
/reload-plugins

Plugin trust ladder (safe order)

1. First-party connectors enabled via Claude UI with explicit OAuth scopes.
2. Official Anthropic marketplace plugins (claude-plugins-official).
3. Reviewed directory or known-vendor marketplaces with documented ownership.
4. Custom remote MCP connectors you control and monitor.
5. Random community scripts/plugins only after code review and least-privilege setup.

Curated, practical list. “Free/open” means fully open-source or free-to-start with optional paid tiers. Use only what you need.

Best free extras to install first (tiered)

These files are high-leverage because they shape Claude behavior, team alignment, and auditability.

Managed policy and organization-level paths

• Managed CLAUDE.md (machine-wide): macOS /Library/Application Support/ClaudeCode/CLAUDE.md, Linux/WSL /etc/claude-code/CLAUDE.md, Windows C:\Program Files\ClaudeCode\CLAUDE.md.
• Managed settings files: same OS directories support managed-settings.json and managed-settings.d/*.json.
• Windows registry managed settings: HKLM\SOFTWARE\Policies\ClaudeCode with Settings JSON value (with lower-priority user-level fallback at HKCU).

Use managed settings for enforceable controls and managed CLAUDE.md for organization-wide guidance text.

Starter template: CLAUDE.md

# Project instructions for Claude

## Purpose
- One sentence on what this repo is for.

## Boundaries
- Do not touch production paths directly.
- Use deploy scripts in `ops/scripts/`.
- Treat runtime artifacts (.db, .env, caches) as non-source.

## Workflow
1. Inspect current state.
2. Apply smallest safe change.
3. Run local verification.
4. Deploy with script.
5. Run health checks.
6. Commit + push with focused message.

## Repo conventions
- Preferred test command: `...`
- Preferred formatter/linter: `...`
- Branching rules: `...`
              - Required docs updates: `README.md`, `CHANGELOG.md`

Starter template: CLAUDE.local.md

# Local-only Claude instructions

## Scope
- This file is local-only and should not be committed.
- Use it for machine or branch-specific preferences.

## Defaults
- Use existing project shell aliases.
- Keep secrets in environment-specific secret stores, not here.

## Workflow
- Run read-only checks before write actions.
- Pause and confirm before external deployments.

Starter template: .claude/rules/01-safety.md

# Safety rules

## Never do
- Push secrets to logs.
- Execute destructive commands without explicit confirmation.

## Do this first
- Validate `claude --version`.
- Confirm `git status` is expected before edits.
- Run one minimal integration action and check output.

## Escalation
- If uncertain, rollback with a small git commit and re-run health checks.

Starter template: README.md

# Project name

## What this does
Short practical summary.

## Quick start
```bash
git clone ...
cd ...
make dev
```

## Structure
- `app/` runtime code
- `ops/` deploy scripts
- `docs/` deep references

## Deploy
```bash
bash ops/scripts/deploy-website.sh
bash ops/scripts/deploy-website.sh --apply
```

## Validation
```bash
bash ops/scripts/healthcheck.sh website
```

MCP (Model Context Protocol) is the open standard Claude Code uses to connect to external tools and data sources. Use it when you are repeatedly copying data into chat from other systems.

• Best for: issue trackers, monitoring, databases, internal APIs, and custom tooling.
• Transports: HTTP (recommended), SSE (deprecated), local stdio process.
• Trust model: only connect servers you trust; prompt-injection risk is explicitly called out in official docs.

Beginner-safe MCP setup flow

# Add remote HTTP MCP server (recommended)
claude mcp add --transport http notion https://mcp.notion.com/mcp

# Add local stdio MCP server example
claude mcp add airtable -- npx -y airtable-mcp-server

# List and inspect servers
claude mcp list
claude mcp get notion

:: Add remote HTTP MCP server
claude mcp add --transport http notion https://mcp.notion.com/mcp

:: Add remote SSE server (legacy/deprecated transport)
claude mcp add --transport sse asana https://mcp.asana.com/sse

:: Manage servers
claude mcp list
claude mcp get notion

# Open MCP status panel in Claude Code
/mcp

# If connector auth is required, complete OAuth from /mcp

# Remove a server if no longer needed
claude mcp remove notion

Practical verified MCP options (from official docs/examples)

MCP scope + config quick notes

• local scope: only you, current project context.
• project scope: shared via project .mcp.json.
• user scope: your account across projects.

Out-of-the-box connectors are integrations available directly in Claude’s connectors experience (web/desktop/mobile), generally enabled through settings and OAuth, without hand-writing MCP CLI config.

Pros/cons vs hand-managed MCP

• Pros: lower setup friction, UI-led auth, easier for non-CLI users.
• Cons: less custom control than raw MCP configs; capabilities vary by connector.

Concrete patterns you can run today. Start simple and add integration complexity only when usage justifies it.

# Daily baseline
claude
# ask: "review modified files, suggest minimal patch, run tests"

# Add MCP integration for issue flow
claude mcp add --transport http issuehub https://example.com/mcp
claude mcp list

# Run release checks
bash ops/scripts/deploy-website.sh
bash ops/scripts/deploy-website.sh --apply

# Plugin workflow with official + demo marketplaces
/plugin install github@claude-plugins-official
/plugin marketplace add anthropics/claude-code
/plugin install commit-commands@anthropics-claude-code
/reload-plugins

# Scope-aware MCP setup example using an official docs endpoint
claude mcp add --scope project --transport http notion https://mcp.notion.com/mcp

This page includes client-side export utilities only (no backend required):

• Full page PDF: top button uses browser print-to-PDF.
• Section PDF: each section has its own export button.
• Section MD: each section exports to a standalone .md file.
• Full page MD: top button exports all sections into one Markdown document.

Official docs change quickly. Use these as your primary references before acting on any connector/MCP claim:

• Claude Code overview (install paths, platform notes, core surfaces)
• Claude Code MCP docs (transports, scopes, commands, trust notes)
• Discover/install plugins (marketplaces, plugin scope, official marketplace)
• Explore the .claude directory (file locations for CLAUDE.md, rules, settings, skills)
• Claude Code settings/configuration (managed settings paths and precedence)
• Memory + CLAUDE.md docs (instruction files and scope model)
• Set up Claude integrations (availability, enablement, privacy constraints)
• Use connectors to extend Claude's capabilities (directory/custom connector behavior)
• Custom connectors via remote MCP (Anthropic-cloud network behavior and firewall requirements)
• Google Workspace connectors (capabilities and limitations)