Claude Code extensions don't really exist. Here's what does.

If you're searching for "Claude Code extensions," you're probably looking for a way to make Claude Code do more than it does out of the box. Fair. But there's no extension store, no plugin marketplace, no VS Code-style sidebar where you install add-ons.

What Claude Code does have is three mechanisms for extending behavior. They're all file-based. None of them require downloading anything from a marketplace.

Skill files

A skill file is a markdown document that tells Claude Code how to handle a specific type of task. You write it, put it in your project, and Claude Code loads it when the task comes up.

For example, you might have a skill file called debug-memory-leak.md that walks Claude through your process for finding memory leaks: check the heap snapshots, look for event listener accumulation, run the profiler, compare before and after. You'd never put this in agents.md because you don't need it every session.

Not sure if your idea should be a skill file or go in agents.md? ClaudeOrNot can tell you.

MCP servers

MCP stands for Model Context Protocol. It's a way to give Claude Code access to external tools and data sources. Think of it as a bridge between Claude and things outside your codebase: databases, APIs, documentation sites, ticket systems.

You configure MCP servers in a JSON file. Claude Code connects to them and can call their tools during a conversation. For instance, you could set up an MCP server that lets Claude query your Postgres database directly instead of you copy-pasting schema dumps.

The ecosystem is still young. Some servers are polished, some are rough. The official Anthropic ones (filesystem, GitHub, Postgres) work reliably. Community ones vary.

Hooks

Hooks let you run custom scripts at specific points in Claude Code's workflow: before it makes a change, after it runs a command, when it starts a session. You can use them to enforce formatting, run linters, or inject context automatically.

This is the closest thing to "extensions" in the traditional sense. But you're writing shell scripts, not installing packages.

So what should you actually do?

If you're just starting out, ignore MCP servers and hooks for now. Start with agents.md and one or two skill files. That's enough to get noticeable improvements in Claude Code's output.

Once you're comfortable with those, add an MCP server for your database or docs. Then experiment with hooks if you want automated quality gates.

The mistake people make is going straight to the fancy stuff and skipping the basics. A well-written agents.md will do more for you than three MCP servers and a hook pipeline.