Best MCP Servers for Claude Code in 2026

The best MCP servers for Claude Code are the few that fill real gaps: current documentation, your code host, a browser, your notes and your design tools. MCP servers are connectors that give Claude tools beyond your files. Add too many and you pay in context and trust; add the right five and Claude Code gets noticeably more capable.

These are the servers I'd install first, with the command for each, plus the question I ask before adding any server: is there a CLI that does this better?

What is an MCP server?#

The Model Context Protocol is an open standard for connecting AI applications to external systems. Its maintainers compare it to a USB-C port for AI applications: one standard connection instead of a custom integration per tool.

An MCP server wraps a system, such as GitHub, a browser or a documentation index, and exposes tools Claude can call. Claude Code, Claude Desktop, Cursor, VS Code and ChatGPT all support MCP, so one server works across them.

How to add an MCP server to Claude Code#

According to the Claude Code MCP documentation, remote servers take one command:

claude mcp add --transport http <name> <url>

Local servers that run on your machine put the launch command after two dashes:

claude mcp add <name> -- <command> [args...]

Three things worth knowing:

• Scope. Servers default to local scope. Add --scope project to share a server with your team through the repository, or --scope user to use it in every project.
• Checking. Run /mcp inside a session to see connected servers, authenticate, and disable the ones you don't need.
• Context. Claude Code loads only tool names at first and fetches full definitions when a tool is used, so a connected server costs little until it's called.

The best MCP servers for Claude Code#

1. Context7: up-to-date documentation#

Models learn libraries from training data that's months old, which is where invented APIs come from. Context7 fetches current, version-specific documentation and code examples and puts them in context.

Setup is one command that authenticates and installs either a skill with its ctx7 CLI or an MCP server:

npx ctx7 setup --claude

Then add "use context7" to prompts that involve a library. A free API key raises the rate limits.

2. GitHub: your repositories and pull requests#

The official GitHub MCP Server lets Claude read issues, review pull requests, check Actions runs and work across repositories. GitHub hosts a remote version, and you can run it locally with Docker and a personal access token:

claude mcp add github -e GITHUB_PERSONAL_ACCESS_TOKEN=$GITHUB_PAT -- docker run -i --rm -e GITHUB_PERSONAL_ACCESS_TOKEN ghcr.io/github/github-mcp-server

Give the token only the permissions it needs, and keep it out of committed files. If you already use the gh CLI, try it first: Claude Code runs it directly, and it adds no tool listings to your context.

3. Playwright: a browser Claude can drive#

Playwright MCP from Microsoft lets Claude open pages, click, fill forms and read results. It works from the page's accessibility tree rather than screenshots, which makes actions faster and more deterministic.

claude mcp add playwright npx @playwright/mcp@latest

Its own documentation makes an honest point: coding agents increasingly prefer a CLI plus skills for browser work, because it avoids loading large tool schemas and page snapshots into context. Use the MCP server for exploratory checks, and scripted Playwright tests for anything you run repeatedly.

4. Your notes: Notion or Obsidian#

If your knowledge lives in Notion, its remote server connects in one line, straight from Anthropic's docs:

claude mcp add --transport http notion https://mcp.notion.com/mcp

If it lives in Obsidian, you may not need a server at all, because Claude Code can work on the vault's files directly. When you do want one, for semantic search or for other clients, I compare the options in Obsidian MCP.

5. Claude Design: from canvas to code#

If you use Claude Design for prototypes and pages, its MCP server lets you create and edit designs from Claude Code and pull them into your codebase:

claude mcp add --scope user --transport http claude-design https://api.anthropic.com/v1/design/mcp

Then run /design-login. I cover the design workflow in Claude design skills.

When a CLI beats an MCP server#

Anthropic's cost guidance is direct about this: tools like gh, aws, gcloud and sentry-cli are more context-efficient than MCP servers because they don't add per-tool listings. Claude Code can run them as ordinary commands.

My rule of thumb:

• Use a CLI when a mature one exists and you already have it authenticated.
• Use an MCP server when there's no good CLI, when you need OAuth, or when the same tools must work in Claude Desktop and other clients.
• Use a skill on top of either, to teach Claude the procedure. My breakdown of skills vs agents vs MCP explains how they fit together.

How to connect MCP servers safely#

An MCP server can read data and take actions on your behalf, so treat adding one like installing software:

1. Connect only servers you trust. The Claude Code docs warn that servers fetching external content can expose you to prompt injection.
2. Prefer official servers from the company that runs the service.
3. Scope tokens tightly. Read-only where possible, one repository or workspace where possible.
4. Keep secrets in environment variables, never in a committed .mcp.json.
5. Prune regularly. Run /mcp and disable anything you haven't used this week.

Where to go next#

Start with Context7 and whichever of GitHub or your notes app you use daily. Add Playwright when you start building web UI. That's enough for most people for months.

If you want to set these up on your own projects with feedback, MCP and tool connections are week three of my Claude Code course. And if you're still getting comfortable with Claude Code itself, begin with how to use Claude Code.