Claude Code Setup plugin: codebase-aware hooks, MCP, skills recommendations

What Claude Code Setup does

Claude Code Setup is a read-only official plugin that inspects a repository and recommends MCP servers, skills, hooks, subagents, and slash commands. It is not a bulk installer, so adoption should follow a deliberate review of scope, privileges, and side effects.

The critical product boundary is that the recommender itself is read-only: it does not rewrite project files. Adoption remains a human decision.

---

What this article covers

This guide is for developers deciding whether a repository needs more Claude Code automation. It combines the official product page, Claude Code plugin docs, the claude-plugins-official README, and independent write-ups in this order: product boundary, install, scope choice, then selective adoption.

• what is automated versus what still requires judgment
• how to install safely
• how to adopt only the highest-value suggestions on a real repo

---

Official starting points

Start from the official Claude Code Setup plugin page.

For packaging, install scopes, and marketplace security, also use:

• plugins documentation
• Discover plugins

---

1. What Claude Code Setup is (a read-only automation recommender)

Claude Code Setup is not a bulk installer that rewrites config for you. It is a codebase-aware recommendation engine that produces a prioritized report.

Inputs and stack signals

According to the official page, inputs include project structure, dependencies, and code patterns. In practice that means signals such as package.json, language files, and directory layout.

Stack-specific examples on the page include:

• Playwright MCP for React projects
• security-reviewer style subagents when authentication code is detected

---

Five categories and default volume

Recommendations span five categories:

1. MCP servers
2. skills
3. hooks
4. subagents
5. slash commands

By default, the plugin surfaces the top 1–2 highest-value items per category. Focused questions such as “what MCP servers should I use?” can expand the list to 3–5 options.

The GitHub plugin README reinforces the same read-only constraint and lists concrete examples:

• context7
• Playwright
• auto-format/lint
• security / performance / a11y reviewers
• /test, /pr-review, /explain Treat these as stack-conditional examples, not a default install set—adopt only when the matching signal exists and the privilege blast radius is acceptable.

---

How to treat suggestions

Reader takeaway: treat suggestions as a ranked shortlist, not a mandatory install set.

The plugin is a strong entry point for cleaning up a messy Claude Code setup, but anything you adopt—especially hooks and MCP servers—can execute with real privileges.

Source: Claude Code Setup, claude-plugins-official / claude-code-setup (as of July 2026).

---

2. Install and first run (marketplace, scope, reload)

Installation is intentionally short. Use this checklist:

1. Start Claude Code and open /plugin → Discover, or run install directly.
2. Run /plugin install claude-code-setup@claude-plugins-official.
3. Choose an install scope: user, project, or local.
4. Run /reload-plugins so skills/MCP components reload in-session.
5. If the marketplace is missing: /plugin marketplace add anthropics/claude-plugins-official.
6. If the plugin is not found: /plugin marketplace update claude-plugins-official.

Choosing a scope

Scope	Where it applies	Best fit
user	All of your projects	Personal baseline across repos—only after local/project trials; wrong choice widens blast radius
project	Shared via .claude/settings.json	Team-shared recommendations—prefer this for shared baselines
local	Current repo only, not shared	Experiments and private trials—use first when risk is unclear

---

Trust boundary and first-run completion

Official docs classify plugins and marketplaces as highly trusted components that can execute with user privileges.

• Install only from trusted sources
• Even on claude-plugins-official, the marketplace README notes that Anthropic does not control every MCP server or file referenced by plugins—verify what you enable

---

First-run completion loop

Independent hands-on posts (for example on Zenn) consistently complete the first-run loop as:

1. install
2. reload
3. natural-language recommender prompts on a real repository

Install alone does not produce recommendations.

Source: plugins docs, Discover plugins, Zenn @shirochan (as of July 2026).

---

3. Category comparison — what to adopt, when, and how far

Terminology confusion is common: installing a plugin is not the same as enabling every automation capability. Bulk-enabling raises privilege and side-effect risk, so use the comparison table below and adopt only categories that clear your auth/scope checks.

Cross-category comparison

Category	What it does	Typical examples	Adoption criteria
MCP Servers	Connect external tools/data	context7 (docs), Playwright (frontend), DB MCP	Is external context needed this session? Is auth scope acceptable?
Skills	Reusable specialized workflows	Plan agent, frontend-design, gen-test style skills	Recurring work, or one-off prompt enough?
Hooks	Automations on tool events	auto-format/lint, block .env edits	Failure side effects; whether blocking is required

---

Subagents and slash commands

Category	What it does	Typical examples	Adoption criteria
Subagents	Separated specialist reviews	security / performance / a11y reviewers	Does the current codebase actually need that review lens?
Slash Commands	Short launches for routines	/test, /pr-review, /explain	Team-shared routine vs personal shortcut

---

Sources for this comparison

Source: Claude Code Setup, GitHub README (as of July 2026).

---

Labeling variance and adjacent plugins

Independent verification also notes a documentation nuance:

• the README lists Slash Commands as the fifth category
• SKILL.md may list Plugins instead Treat category labels as potentially drifting documentation details—verify against current docs before treating names as a stable contract.

Treat this as category labeling variance, not product failure. For readers, the useful model is simply: five automation surfaces, recommended from repository context.

---

Adjacent official plugins

Classmethod’s marketplace survey places claude-code-setup among Claude Code configuration/management plugins.

• Adjacent official plugins (Code Review, Security Guidance, Commit Commands, and others) are better added later as gap-fillers
• Prefer that over installing all of them at once

Source: DevelopersIO marketplace survey (survey dated March 2026; listings and counts can change).

---

4. How to use it — natural-language prompts and a 3-phase flow

After install, the main interaction is conversational, not memorizing a private slash command.

Recommended prompts

Official example prompts include (re-check the product page if wording changes; treat them as starting points, not auto-apply scripts):

• recommend automations for this project
• help me set up Claude Code
• what hooks should I use?
• category-specific: what MCP servers should I use?

---

Three-phase internal flow

Independent hands-on analysis describes the recommender roughly as three phases:

1. Scan: use Read/Glob/Grep/Bash-style inspection of package.json, structure, and existing config
2. Match: map stack signals to known patterns (Prettier → format hook, React → Playwright MCP, Postgres → DB MCP, etc.)
3. Report: output reasons and install/implementation steps without auto-applying changes—you still must verify privileges, auth scope, and rollback before enabling anything

---

Practical adoption pattern

That design keeps investigation automated and change control human.

A practical next action—assuming suggestions are candidates, not quality guarantees—is:

• adopt only the top priorities manually
• park the rest as backlog
• expect thinner repositories with sparse config to receive more generic suggestions

---

Why start on a real repository

Coverage on note.com similarly frames the plugin as a five-category scan that reduces the upfront cost of manual catalog research.

• Running the recommender once on a real repo is usually a better first lesson
• than reading the full marketplace catalog

Source: Claude Code Setup, Zenn @shirochan, note @hacklog_stealth (as of July 2026).

---

5. Adoption checklist and safety boundaries

A recommendation is not a quality guarantee. Suggestion quality depends on repository hygiene, and recommended ≠ apply immediately. Treat sparse config as a risk that outputs become generic, and verify evidence, auth scope, and rollback before enabling anything.

Pre-adoption checklist

Use this minimum checklist before enabling anything:

• ✓ Does the stated evidence (detected deps/paths) match your understanding?
• ✓ Are auth, privileges, and network access for MCP/hooks acceptable?
• ✓ Should this live in user scope for every machine, or stay project/local?
• ✓ Can you start with only 1–2 items and verify side effects after /reload-plugins?
• ✓ Are you mixing untrusted third-party marketplace plugins?
• ✓ Do you have a rollback path (disable plugin/components, review config diffs)?

---

Boundaries that matter most

Three boundaries matter most:

1. The Setup plugin is read-only, but adopted hooks/MCP components can execute with privileges.
2. Sparse repositories tend to produce generic recommendations.
3. Install counts reported by third-party catalogs (for example Composio’s order-of-magnitude scale notes) are popularity signals, not quality certification.

Read these together with the official security guidance: “install from trusted marketplaces” is necessary but not sufficient. You still need to review what each suggestion can touch.

---

Team scope staging

In team settings, stage the scopes:

• prefer project scope for shared baselines (limits accidental machine-wide bleed)
• local for experiments (contains blast radius until side effects are known)
• user for true personal defaults only after local/project trials succeed

Source: plugins docs, Composio Claude Code plugins overview (as of July 2026).

---

6. Author Perspective

Claude Code write-ups often bias toward installing every convenient extension. A cleaner review order is the opposite: narrow suggestions first and freeze privilege boundaries early.

Claude Code Setup’s main value is not exhaustive coverage; it is a fast, repository-aware ranking of what is worth reviewing next.

• Treat verified marketplace placement and install scale as entry signals
• Then decide from dependency evidence, auth scope, and the cost of undoing a change

---

FAQ

Q1. Does Claude Code Setup rewrite config files automatically?

No. The official product page and README both describe a read-only recommender. You apply changes yourself.

---

Q2. What is the install command?

/plugin install claude-code-setup@claude-plugins-official, then /reload-plugins. If the marketplace is missing or stale, use the official marketplace add / marketplace update flow first.

---

Q3. How many recommendations appear?

Defaults are top 1–2 per category. Focused category questions can expand to 3–5 options (official plugin page).

---

Q4. How do plugins, skills, MCP servers, and hooks differ?

A plugin is the package/distribution unit. Skills, hooks, MCP servers, and subagents are capabilities that can ship inside a package. Claude Code Setup is the repository-aware recommender entry point for those capabilities.

---

Q5. What related topics should I read next?

For structured learning, see the Anthropic Academy free courses. For long-running agent loops and stop conditions, see Claude Code loop design. For agent-building OSS scaffolding, see Launch Your Agent.

---

Summary — what to do next

1. Install Claude Code Setup from the official marketplace and run /reload-plugins.
2. On a real repository, run one recommender prompt such as recommend automations for this project.
3. Use the category table and checklist to adopt only top priorities, then verify side effects.
4. Evaluate related official plugins later as gap-fillers, not as a bulk install list.

Claude Code Setup is a read-only recommender for turning a messy Claude Code setup into a production-minded baseline. Do not treat it as auto-write tooling. Use it as a prioritized adoption queue, and the value becomes both faster and safer.

Related articles:

• Free Anthropic Academy courses for Claude Code, MCP, and Agent Skills
• Claude Code Loop Design: Turn, Goal, Time, Proactive Patterns
• Anthropic launch-your-agent Claude Code Skill Open Source — Build Managed Agents Fast