المهارات وبروتوكولات MCP

Out of the box, Claude Code can read files, edit code, and run shell commands. But the real power comes when you extend it — with Skills that teach it your team's patterns, and MCP servers that connect it to external systems.

Skills: teaching Claude your patterns

A Skill is a markdown file that Claude applies automatically when it detects a matching task. You don't call a skill explicitly — Claude reads the skill's description and decides when to use it.

Skills live in three places:

• Project skills: .claude/skills/ in your repo — shared with the team via git
• User skills: ~/.claude/skills/ — personal, across all projects
• Anthropic skills: built into Claude Code (like /commit, /review-pr)

Here's what a skill file looks like:

---
description: "Enforce our commit message format: type(scope): description"
---

When creating a commit message, follow this format exactly:
- Type: feat, fix, refactor, docs, test, chore
- Scope: the component or area (e.g., auth, api, ui)
- Description: imperative, lowercase, no period
- Body: explain WHY, not what

Example: feat(auth): add magic link login flow

When you run /commit, Claude reads all available skills. If your commit-format skill matches, Claude follows its instructions without you asking.

When to create a skill:

• You find yourself giving Claude the same instruction repeatedly
• Your team has a convention Claude doesn't follow by default
• You want a repeatable workflow (deploy checklist, PR review criteria)

MCP servers: connecting to external systems

The Model Context Protocol is an open standard that lets AI models talk to external tools and data sources. An MCP server is a program that exposes tools (actions), resources (data), and prompts (templates) to Claude.

Claude Code ships with MCP support built in. To add a server, edit .claude/settings.json:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@anthropic-ai/mcp-playwright"]
    }
  }
}

Once configured, Claude Code discovers the server's capabilities automatically. You'll see new tools available — like browser_navigate, browser_screenshot, browser_click — and Claude will use them when the task calls for it.

Popular MCP servers you should know about:

Server	What it does
Playwright	Browser automation — navigate, click, screenshot, test
Postgres	Query your database directly from Claude
GitHub	Access repos, issues, PRs without gh CLI
Context7	Fetch live documentation for any library
Filesystem	Read/write files outside the project directory

The MCP ecosystem has over 34,000 npm-dependent projects as of April 2026 and is growing fast. The 2026 roadmap focuses on transport scalability, enterprise auth, and agent-to-agent communication.

Skills vs MCPs: when to use which

Use a Skill when...	Use an MCP when...
You want to teach Claude a pattern	You want to connect Claude to an external system
The instruction is about how to do something	The capability is about what Claude can access
It's project- or team-specific knowledge	It's a general-purpose tool any project could use
Example: commit format, code review checklist	Example: database access, browser control, API calls

They compose well together. A skill might say "when deploying, run the Playwright test suite first" — and an MCP server provides the Playwright tools that make that possible.

What's next

In the next lesson, you'll learn about subagents — spawning parallel Claude instances that work on different parts of a task without polluting each other's context. It's the feature that makes Claude Code scale from simple tasks to complex, multi-file projects.