Claude Official Plugins for development

Claude Code’s official plugin ecosystem is not only a list of Anthropic-built tools. It also contains partner plugins, external service integrations, and third-party entries.

This article is narrower: it covers the plugin directories under plugins/ in Anthropic’s official claude-plugins-official GitHub repository. I checked that directory on 20 July 2026 and use it here as the scope for internal Anthropic-maintained plugins.

For a Thailand business website, villa booking system, WordPress install, Laravel backend, or Astro frontend, the practical question is not “how many plugins can I install?” It is: which plugin adds useful context without making Claude Code heavier, noisier, or harder to review?

What counts as official here

The sources I used were the Claude plugin directory, the official Claude Code plugin discovery documentation, the official plugins/ directory, and the public claude-plugins-official marketplace catalog.

This scope includes 38 plugin directories. It excludes external_plugins/, partner-maintained entries, the community marketplace, demo marketplaces, and external service connectors such as GitHub, GitLab, Linear, Asana, Firebase, Playwright, and similar integrations. One of the 38, example-plugin, is a reference plugin for plugin authors rather than a normal production workflow plugin.

The normal install format is:

/plugin install plugin-name@claude-plugins-official

After installation, run /reload-plugins if you want the current Claude Code session to pick up the change without restarting.

Code intelligence plugins

These plugins connect Claude Code to language servers. They help Claude see diagnostics after edits and navigate code with language-aware tools instead of relying only on text search.

They do not replace tests, type checks, or a developer review. They make Claude’s local feedback loop better when the matching language server is installed on the machine.

Plugin	Language or toolchain	What it adds
clangd-lsp	C and C++	Code intelligence through clangd: diagnostics, symbols, definitions, references, and type-aware navigation.
csharp-lsp	C#	C# language-server support so Claude can inspect errors and navigate .NET code more precisely.
gopls-lsp	Go	Go code intelligence and refactoring context through gopls.
jdtls-lsp	Java	Eclipse JDT.LS support for Java diagnostics and navigation.
kotlin-lsp	Kotlin	Kotlin language-server support for Kotlin projects.
lua-lsp	Lua	Lua language-server support for projects with Lua scripts or embedded Lua code.
php-lsp	PHP	Intelephense-backed PHP code intelligence, useful for WordPress, Laravel, and older PHP codebases.
pyright-lsp	Python	Pyright diagnostics, type checking, and navigation for Python projects.
ruby-lsp	Ruby	Ruby language-server analysis and navigation.
rust-analyzer-lsp	Rust	Rust Analyzer diagnostics and navigation for Rust projects.
swift-lsp	Swift	SourceKit-LSP support for Swift codebases.
typescript-lsp	TypeScript and JavaScript	TypeScript language-server support for .ts, .tsx, .js, .jsx, and related module formats.

For the kind of website work I usually see in Thailand, php-lsp and typescript-lsp are the obvious starting points. They help when Claude is working around custom WordPress plugin code, Laravel controllers, Astro components, frontend TypeScript, or API integration code.

Development workflow plugins

These plugins shape how Claude Code plans, builds, modernizes, and extends software.

Plugin	What it does
agent-sdk-dev	Adds development help for working with the Claude Agent SDK. Use it when building or debugging agent-based applications.
code-modernization	Guides legacy modernization work through a structured workflow. It is aimed at larger transformations such as old Java, C++, COBOL, or monolith web applications.
feature-dev	Adds a fuller feature-development workflow with agents for exploration, architecture, and quality review. Useful when a feature touches several parts of a codebase.
frontend-design	Helps Claude produce more deliberate frontend interfaces instead of generic AI-looking layouts. Useful for UI work, but still needs design review.
mcp-server-dev	Helps design and build MCP servers, including tool design, authentication, deployment models, and interactive MCP apps.
mcp-tunnels	Helps connect Claude to private MCP servers through an Anthropic MCP tunnel, with a Docker Compose based setup flow.
playground	Creates self-contained interactive HTML playgrounds with controls, live preview, and copyable prompt output. Useful for exploring UI ideas, data, and concepts.
plugin-dev	Provides a toolkit for creating Claude Code plugins, including hooks, MCP integration, commands, agents, and validation practices.
ralph-loop	Runs repeated self-referential passes on the same task so Claude can inspect and improve its previous work. Use carefully, because repetition is not a substitute for verification.
skill-creator	Helps create, improve, test, and benchmark Claude skills. Useful when project-specific workflows are worth packaging as durable instructions.

For smaller business websites, I would not start by installing all of these. I would start with the problem: are you improving a UI, creating an MCP server, maintaining skills, or modernizing an old codebase? Install the plugin that matches that job.

Review, maintenance, and project memory

These plugins are closer to everyday repository hygiene.

Plugin	What it does
claude-code-setup	Analyzes a codebase and recommends useful Claude Code automations such as hooks, skills, MCP servers, and subagents.
claude-md-management	Helps maintain CLAUDE.md: audits quality, captures session learnings, and keeps project memory current.
code-review	Runs automated code review for pull requests with multiple specialized agents and confidence scoring.
code-simplifier	Focuses on simplifying recently modified code while preserving behavior. Useful after a rough implementation pass.
commit-commands	Adds Git commit workflow commands, including commit, push, and pull request creation.
hookify	Helps create custom hooks from repeated conversation patterns or explicit instructions. Useful when Claude keeps making the same avoidable mistake.
pr-review-toolkit	Adds specialized PR review agents for comments, tests, error handling, type design, code quality, and simplification.
project-artifact	Generates a living project status page as a shareable Claude artifact, backed by project configuration and refreshed from current state.
receipts	Generates a local impact report from Claude Code session transcripts, summarising work by project and relating usage to files and commits without inventing cost or time-saved figures.
session-report	Creates an HTML report from local Claude Code transcripts, including token use, cache efficiency, subagents, skills, and expensive prompts.

These are the plugins I would look at first for an existing Thailand-focused codebase. A villa booking flow, a lead form, a WordPress checkout, or a multilingual Astro site benefits more from careful review, memory, and diagnostics than from broad automation.

I wrote more about that workflow in Claude Code for existing codebases in Thailand, AGENTS.md and CLAUDE.md for AI coding instructions, and AI code review with Claude and Codex.

Security and safety

Plugin	What it does
security-guidance	Reviews Claude-generated code for common security issues. It can warn on edits, review diffs, and inspect commits for problems such as injection, XSS, SSRF, hardcoded secrets, and other vulnerability classes.

This is one of the most practical internal plugins for business websites. It does not make a change safe by itself, but it adds a useful review pass while Claude is writing code.

For production systems, especially contact forms, booking flows, payment callbacks, authentication, and admin features, security review should still be paired with ordinary engineering checks: diff review, tests, logs, and staging.

Output styles and learning

Plugin	What it does
explanatory-output-style	Adds more educational explanation around implementation choices and codebase patterns. Useful when you want to understand why a change is being made.
learning-output-style	Turns the session into a more interactive learning mode where Claude asks for meaningful code contributions at decision points.

These are not repository automation plugins. They change how Claude communicates. That can be useful when you are learning a codebase, onboarding someone, or slowing down enough to understand a risky change.

Specialized plugins

Plugin	What it does
cwc-makers	Helps onboard a Code-with-Claude Makers Cardputer using a setup command that clones the relevant repo, flashes firmware, and installs the Claude Buddy app bundle.
example-plugin	Demonstrates Claude Code extension options, including skills, slash commands, and MCP server configuration. Treat it as a reference for plugin authors, not a normal website workflow plugin.
math-olympiad	Helps solve competition math problems with adversarial verification and calibrated abstention. It is not a normal web development plugin, but it is part of the Anthropic-authored official set.

These are narrow tools. They should only be installed when the task actually matches them.

A practical starting set

For a small or medium website codebase, I would keep the plugin set boring:

• typescript-lsp if the project has TypeScript, Astro, React, Vue, or modern JavaScript
• php-lsp if the project has WordPress, Laravel, or custom PHP
• security-guidance for code that touches forms, payments, auth, APIs, or admin areas
• commit-commands if Claude Code is helping with Git workflow
• code-review or pr-review-toolkit when you want an extra review pass
• claude-md-management when project memory matters across repeated sessions
• claude-code-setup when you are not sure which automations are worth adding

That is enough for many real projects. A codebase like the Thailand Villas case does not need every possible AI automation. It needs changes that are scoped, reviewable, and safe for the business workflow.

What to avoid

Do not treat the official marketplace as a checklist. Each enabled plugin can add context, tools, hooks, or behavior. More plugins can mean more capability, but also more moving parts.

Before installing a plugin, ask:

• Does this solve a current problem?
• Does it need local binaries, credentials, or external access?
• Will it change Claude’s behavior in every session?
• Can I verify the result without trusting the plugin blindly?
• Should it be installed globally, only in this repo, or only for one local session?

For commercial website work, especially on existing systems, the useful path is controlled adoption. Add one plugin, understand what it contributes, verify it on a real task, then decide whether it belongs in the normal workflow.

Frequently Asked Questions

Does this guide cover every plugin in the Claude plugin marketplace?

No. It only covers the internal Anthropic-maintained plugins under the plugins/ directory in the official claude-plugins-official repository as of 20 July 2026. It excludes external_plugins/, partner-maintained entries, the community marketplace, and connectors such as GitHub, GitLab, or Linear.

Should a small business codebase install every available plugin?

No. A boring, targeted set is usually enough: a matching language-server plugin, security-guidance for forms and payments, and something like commit-commands or code-review if it fits the workflow. Each installed plugin adds context, tools, or behavior that has to be understood and verified.

Do the code intelligence plugins replace tests and type checks?

No. Plugins such as php-lsp or typescript-lsp connect Claude Code to a language server so it can see diagnostics and navigate code more precisely. They improve the local feedback loop but do not replace tests, type checks, or a developer review.

Does the security-guidance plugin make AI-generated code safe by itself?

No. It reviews Claude-generated code for common issues such as injection, XSS, SSRF, and hardcoded secrets, and adds a useful review pass. For production systems like contact forms or payment callbacks, it should still be paired with ordinary engineering checks: diff review, tests, logs, and staging.

When I can help

I can help set up Claude Code workflows for existing PHP, WordPress, Laravel, Astro, and API-heavy projects connected to Thailand businesses. That can include choosing practical plugins, writing CLAUDE.md or shared AI instructions, adding review rules, and keeping AI-assisted changes small enough to inspect.

If you want to use Claude Code on a real business codebase without turning the repo into an experiment, send me the project context and I can suggest a careful starting setup.