Don't treat Claude Code as a plugin! Four-layer architecture: built-in, MCP, Plugins, Skills—clarify each once.

The latest list conflates the built-in features of Claude Code, MCP, and plugins. In fact, its expansion is four-layered; understanding the native capabilities and connecting externally as needed is the key to unleashing its power.

Recently, a list titled "9 Plugins That Make Claude Code Like a Veteran Engineer" has been circulating in the community, listing tools like Context7, GitHub, Playwright, terminal, memory, etc. The idea is good, but calling all "built-in features, MCP server, plugins" as plugins can easily mislead beginners. Actually, Claude Code’s expansion is layered, and understanding which layer solves what problem is more important than installing a bunch of things.

Claude Code’s expansion is actually divided into four layers

The first layer is built-in tools, ready to use out of the box, no installation needed, including executing terminal commands (Bash), reading and writing project files, code search, web scraping, extended thinking, and using CLAUDE.md with memory mechanisms to preserve project context. The second layer is MCP (Model Context Protocol) server, which connects via standard protocols through connectors, allowing Claude to access external data and services, such as latest documents, GitHub, browser automation, or databases. This is truly the part that needs to be "installed separately." The third layer is Skills, which package a set of commands and code into on-demand loadable skills, like a docx skill for generating Word documents. The fourth layer is Plugins, which bundle Skills, slash commands, subagents, hooks, and MCP settings into one package for easy installation and sharing.

In that list, actually half of the items don’t need to be installed at all

The list claims that terminal access, file system, memory, and sequential thinking require plugins, but these are already built into Claude Code. Statements like "no file access, AI is useless" or "once Claude gets the terminal" do not apply to Claude Code, because it can execute commands, read/write entire projects, save memories, and reason step-by-step natively. The table below restores those 9 items to their true identities:

| List Item | Actual Function | | --- | --- | | Context7 | MCP server (fetching latest documents) | | GitHub | MCP server | | Playwright | MCP server (browser automation testing) | | Database | MCP server | | Terminal | Built-in (Bash commands) | | Filesystem | Built-in (native project file read/write) | | Memory | Built-in (CLAUDE.md/memory mechanism) | | Sequential Thinking | Has an MCP of the same name, overlapping with built-in extended thinking | | Browser Tools | Built-in web scraping, with optional MCPs like Playwright |

In other words, only about four of these are truly "external," the rest are capabilities Claude Code already has natively.

The MCP servers worth connecting externally

The capabilities that are not built-in but emerge after external connection mainly depend on MCP servers. Context7 feeds Claude the latest framework and library documents, reducing the use of outdated APIs; GitHub MCP allows it to read repositories, PRs, and issues, understanding the entire project structure; Playwright MCP can actually open browsers, click buttons, reproduce, and verify frontend issues; database MCP can query schemas, run queries, and optimize SQL. To add these, simply register the corresponding MCP servers in the configuration file. For details, refer to the complete MCP guide.

The key is not to install the most, but to clarify what each layer solves

The core idea of that list is actually correct: Claude Code’s true power comes from high-context tool integration, not just better prompting. But calling everything a plugin can make people think they need to install a bunch of things to use it, which causes them to overlook its native capabilities. First, clearly distinguish what problems are solved by built-in features, MCP, Skills, and Plugins. Then connect externally as needed. This is the correct order to turn Claude Code from a chat tool into a reliable engineering teammate.

• This article is reprinted with permission from: 《Chain News》
• Original title: 《Don’t Mistake Claude Code as Just Plugins: Clarify Built-in, MCP, Plugins, Skills All at Once》
• Original author: Elponcrab