TL;DR — MCP (Model Context Protocol) is an open standard that gives Claude Code direct connections to external tools. Five servers cover 80% of daily needs: GitHub for PR management, Brave Search for real-time docs lookup, Playwright for browser automation, Filesystem for cross-directory access, and Fetch for pulling web content. Each installs in one command. Jump to the setup →

You’re debugging a production issue. You need to check the related GitHub PR, search for a Stack Overflow answer, and read an API doc (all at the same time). With plain Claude Code, you spend half your time copy-pasting between browser tabs. With MCP servers, Claude does it all without leaving your terminal.

That’s the promise of MCP. Here are the 5 servers that make it real.

What Is MCP and Why Does It Matter?

Model Context Protocol is an open standard that lets Claude Code connect to external tools and data sources. Think of it as USB ports for your AI assistant. Each MCP server is a peripheral that gives Claude a new capability.

The result: Claude stops being a code editor and starts being a connected development environment.

Key insight: Model Context Protocol is an open standard, meaning MCP servers built for Claude Code work with any MCP-compatible client. Anthropic published the spec and maintains the reference server repository at github.com/modelcontextprotocol/servers, but any developer can build and publish a compliant server. The ecosystem already includes hundreds of community-maintained servers beyond the five covered here.

How Do You Add MCP Servers to Claude Code?

Two transport types. Choose based on where the server runs:

Terminal window
# Remote servers (hosted on the web)
claude mcp add --transport http <name> <url>


# Local servers (run on your machine via npx/binary)
claude mcp add --transport stdio [--env KEY=val] <name> -- <command> [args...]

Key things to know:

  • Options come before the server name
  • -- separates the server name from the command that runs it
  • Default scope is local (current project only). Use --scope user for all projects.
Terminal window
# Check what you have installed
claude mcp list


# Remove a server
claude mcp remove <name>


# Inside an active session, check status
/mcp

Now let’s set them up.

1. GitHub MCP — How Do You Manage PRs and Issues Without Leaving Terminal?

What it does: Create PRs, review issues, search code, and manage repos, all through Claude Code. No browser tab switching.

Install:

Terminal window
claude mcp add --transport http github https://api.githubcopilot.com/mcp/

Authenticate: Start a Claude Code session, then run /mcp. It will prompt you to authenticate with GitHub.

Use case: You’re writing a fix and want Claude to check if there’s already an open issue for it, then create a PR with the right labels once you’re done. With GitHub MCP, that’s a single conversation.

Important: The old @modelcontextprotocol/server-github npm package is archived. Use this remote server instead.

Key insight: The GitHub MCP server uses HTTP transport and authenticates via GitHub’s standard OAuth flow — no API token to generate or rotate manually. It exposes tools for creating PRs, searching code across repos, listing issues, and managing labels, all accessible from Claude Code without leaving the terminal or opening a browser tab.

2. Brave Search MCP — How Do You Get Real-Time Web Search in Claude Code?

What it does: Real-time web search inside Claude Code. Find documentation, check Stack Overflow, look up current package versions.

Prerequisites: Free API key from brave.com/search/api. The free tier gives you 2,000 queries/month (plenty for development use).

Install:

Terminal window
claude mcp add --transport stdio \
  --env BRAVE_API_KEY=YOUR_KEY \
  brave-search -- \
  npx -y @brave/brave-search-mcp-server

Replace YOUR_KEY with your actual Brave Search API key.

Use case: You’re implementing OAuth and want Claude to search for current best practices. Instead of pausing to Google it, Claude searches directly and incorporates the results into its response.

Important: The old @modelcontextprotocol/server-brave-search is archived. Use @brave/brave-search-mcp-server from Brave directly.

Key insight: Brave Search’s free API tier provides 2,000 queries per month with no credit card required — sufficient for typical development use. Unlike web scraping approaches, it returns structured search results with titles, URLs, and descriptions that Claude can reason over directly, rather than raw HTML that consumes context window tokens inefficiently.

Try it now: Install the Brave Search MCP server with your free API key and ask Claude: “Search for the latest breaking changes in [a library you use].” Compare how much faster this is than opening a browser tab and parsing the results yourself.

3. Playwright MCP — Browser Automation Built In

What it does: Control a real browser from Claude Code. Take screenshots, fill forms, run E2E tests, scrape content. Microsoft maintains this one.

Install:

Terminal window
claude mcp add --transport stdio playwright -- npx -y @playwright/mcp@latest

No API keys needed.

Use case: You’ve built a form and want to verify it actually works end-to-end. Ask Claude to open the browser, fill in the fields, submit, and screenshot the result. Done in one prompt. No manual testing, no writing a test script first.

Also useful for debugging visual issues. “Screenshot what my login page looks like on mobile” is a valid prompt when Playwright MCP is connected.

Key insight: Microsoft’s Playwright MCP server uses stdio transport and requires no API keys — it launches a real Chromium browser on your machine. This makes it suitable for E2E testing workflows where Claude fills forms, clicks buttons, and verifies visual state, replacing manual QA steps that previously required writing a separate test script before you could verify behavior.

4. Filesystem MCP — Controlled Access Outside Your Project

What it does: Gives Claude read/write access to directories you specify. Claude Code already has access to your project directory. This server extends that to other paths on your system.

Install:

Terminal window
claude mcp add --transport stdio filesystem -- \
  npx -y @modelcontextprotocol/server-filesystem /path/to/allowed/dir

Replace /path/to/allowed/dir with the actual path. You can specify multiple directories by adding more paths at the end.

Use case: Your project reads config files from ~/configs/ or writes output to ~/Documents/reports/. Claude needs to read those files to help you debug. Without Filesystem MCP, you’d have to copy-paste file contents manually. With it, Claude reads them directly.

Keep the allowed paths narrow. Don’t grant access to your entire home directory, give it only what Claude needs.

Key insight: The Filesystem MCP server enforces path-based access control at the server level — Claude cannot read or write outside the directories you specify at install time, regardless of what it is asked to do. This makes it safe to grant access to config directories or shared workspaces without exposing your entire home directory to the session.

5. Fetch MCP — Pull Web Content Into Your Conversation

What it does: Fetch any URL and convert it to clean, readable markdown. Great for pulling in documentation, reading changelogs, or analyzing web content.

Install:

Terminal window
claude mcp add --transport stdio fetch -- npx -y @modelcontextprotocol/server-fetch

Use case: You’re integrating a third-party API and the docs are on a website, not in your codebase. Ask Claude to fetch the docs page and write the integration based on what it reads. No copy-pasting, no context window stuffing.

Also useful for: reading release notes before upgrading a package, pulling in spec sheets, or grabbing any reference material that lives on the web.

Key insight: The Fetch MCP server converts web pages to clean Markdown before passing them to Claude, stripping navigation, ads, and boilerplate that would waste context window tokens. For documentation-heavy integrations — OAuth flows, payment APIs, third-party webhooks — this means Claude reads the same reference material a developer would, without requiring manual copy-paste from the browser.

MCP Server Cheat Sheet

ServerTransportInstall CommandBest For
GitHubHTTPclaude mcp add --transport http github https://api.githubcopilot.com/mcp/PRs, issues, code search
Brave Searchstdioclaude mcp add --transport stdio --env BRAVE_API_KEY=KEY brave-search -- npx -y @brave/brave-search-mcp-serverWeb search, docs lookup
Playwrightstdioclaude mcp add --transport stdio playwright -- npx -y @playwright/mcp@latestBrowser automation, screenshots
Filesystemstdioclaude mcp add --transport stdio filesystem -- npx -y @modelcontextprotocol/server-filesystem /your/pathFiles outside project dir
Fetchstdioclaude mcp add --transport stdio fetch -- npx -y @modelcontextprotocol/server-fetchFetch URLs as markdown

Which MCP Packages Are Archived and Out of Date?

The MCP ecosystem moves fast. A lot of the original @modelcontextprotocol/ npm packages have been archived, including the ones for GitHub, Brave Search, PostgreSQL, SQLite, and Puppeteer.

Before installing any MCP server, check the official MCP servers repository for the current maintained version. The archived packages still install, but they won’t receive updates or bug fixes.

The 5 servers in this guide are current as of March 2026.

Key insight: The original @modelcontextprotocol/ npm packages for GitHub, Brave Search, PostgreSQL, SQLite, and Puppeteer have all been archived. The MCP ecosystem moved quickly from centrally-maintained reference implementations to vendor-maintained packages — GitHub’s server now lives at github.com/github/github-mcp-server, Brave’s at @brave/brave-search-mcp-server. Always check the official servers repository before installing any MCP package.

Get weekly Claude Code tips — One practical tip every week. No fluff, no spam. Subscribe to AI Developer Weekly →

Which MCP Servers Should You Start With?

Don’t install all 5 at once. Start with the two you’ll use most:

  • Working heavily with GitHub? Add GitHub MCP first.
  • Constantly searching for docs? Add Brave Search first.

Once those feel natural, add the rest. Each server you add makes Claude Code more capable, but only if you actually use it.

These 5 servers cover 80% of what you’ll reach for daily. Web search, GitHub, browser control, file access, URL fetching. Once they’re set up, Claude Code stops being a smart text editor and starts being a connected development environment that reaches across your entire stack.

That’s the shift worth making. And with hundreds of community-maintained MCP servers already in existence, it’s one reason the argument that Claude Code’s ecosystem is starting to look like WordPress keeps coming up — the “free core, paid ecosystem” pattern is already taking shape.