Best MCP Servers for Claude Code (2026): 10 Picks

There are thousands of MCP servers, and most of them you’ll never need. This list is the short version: ten servers that pay off for day-to-day software work in Claude Code, each with the install command from its official docs and a note on when it’s worth the connection.

Every command below was checked against the server’s README or vendor docs and the Claude Code MCP documentation in October 2026. New to MCP? Read our MCP servers guide first.

The list at a glance

#	Server	Best for	Maintained by	Transport	Auth
1	GitHub MCP Server	Issues, PRs, code search	GitHub	HTTP (remote) or Docker	PAT or OAuth
2	Context7	Current library docs	Upstash	HTTP (remote)	API key
3	Playwright MCP	Driving a browser	Microsoft	stdio (npx)	None
4	Chrome DevTools MCP	Performance, console, network	Chrome DevTools team	stdio (npx)	None
5	DBHub	Querying SQL databases	Bytebase	stdio (npx)	DSN
6	Sentry	Production errors	Sentry	HTTP (remote)	OAuth
7	Linear	Issue tracking	Linear	HTTP (remote)	OAuth
8	Notion	Specs and team docs	Notion	HTTP (remote)	OAuth
9	Figma	Design-to-code	Figma	HTTP (remote)	OAuth
10	Reference servers	Fetch, Git, Filesystem, Memory	MCP steering group	stdio	None

Bonus: MCP Inspector isn’t a server you give Claude, but it’s the tool to reach for when one of the above won’t connect.

1. GitHub MCP Server

GitHub’s own server gives Claude structured access to repositories, issues, pull requests and reviews. It replaced the archived GitHub server from the reference repo.

claude mcp add --transport http github https://api.githubcopilot.com/mcp/ \
  --header "Authorization: Bearer YOUR_GITHUB_PAT"

Use it when you triage issues, review PRs or search code across repos. Claude Code can already run git and gh, so skip it if the CLI covers your needs. Use a fine-grained token limited to the repos Claude should touch. Details: GitHub MCP Server.

2. Context7

Context7 feeds Claude current, version-specific documentation for libraries and frameworks, which fixes the most common cause of wrong code: APIs that changed after the model’s training data.

npx ctx7 setup --claude

The setup CLI signs you in and lets you choose MCP mode or a CLI-plus-skill mode. Manual MCP setup:

claude mcp add --transport http context7 https://mcp.context7.com/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Use it when you work with fast-moving frameworks or upgrade between major versions. Details: Context7.

3. Playwright MCP

Microsoft’s browser automation server. Claude opens pages, clicks, types and reads results, working mainly from the page’s accessibility snapshot instead of screenshots.

claude mcp add playwright npx @playwright/mcp@latest

Use it when Claude needs to check the UI it just built or reproduce a click-path bug. Needs Node.js 18 or newer. Stick to sites you trust: page content can carry prompt injection. Details: Playwright MCP.

4. Chrome DevTools MCP

From the Chrome DevTools team. It connects Claude to a live Chrome with DevTools capabilities: performance traces, console messages and network requests.

claude mcp add chrome-devtools -- npx -y chrome-devtools-mcp@latest

It’s also available as a plugin from its own marketplace (/plugin marketplace add ChromeDevTools/chrome-devtools-mcp). Use it when the question is “why is this slow” or “which request failed”, not just “does this button work”. Details: Chrome DevTools MCP.

5. DBHub

Bytebase’s DBHub connects Claude to PostgreSQL, MySQL, SQL Server, Oracle, MariaDB or SQLite through a connection string. It’s the example the Claude Code docs use for database queries.

claude mcp add --transport stdio db -- npx -y @bytebase/dbhub \
  --dsn "postgresql://readonly:pass@host:5432/analytics"

Use it when you explore an unfamiliar schema or answer ad-hoc data questions. Always use a read-only database user: the database permissions are your real safety boundary, not the prompt. Details: DBHub.

6. Sentry

Sentry’s hosted MCP server lets Claude search errors, analyse performance and triage issues in your Sentry organisation. All connections use OAuth.

claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

Then run /mcp inside Claude Code, select sentry and choose Authenticate to finish the browser login. Sentry’s docs also show URLs scoped to one organisation and project:

claude mcp add --transport http sentry https://mcp.sentry.dev/mcp/your-org/your-project

Use it when you debug production errors: “find the top unresolved error from the last release and propose a fix” becomes one prompt that spans Sentry and your code.

7. Linear

Linear’s remote server can find, create and update issues, projects and comments.

claude mcp add --transport http linear-server https://mcp.linear.app/mcp

Authentication is OAuth by default; Linear also documents bearer tokens and API keys. Use it when your backlog lives in Linear and you want Claude to turn a debugging session into a ticket, or pick up a ticket and implement it.

8. Notion

Notion’s hosted server lets Claude search pages, create and update documents and work with databases.

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

Run /mcp to authenticate. Use it when specs, runbooks or decision records live in Notion and you’re tired of pasting them into chat.

9. Figma

Figma’s remote MCP server gives Claude access to design files, component information and design tokens so it can translate designs into code. Figma’s docs recommend the plugin route:

claude plugin install figma@claude-plugins-official

Or add the server directly:

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

Use it when you implement designs and want Claude to read real spacing, colours and component names instead of guessing from a screenshot.

10. The MCP reference servers

The modelcontextprotocol/servers repo holds a small set of reference servers: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time. Several are handy day to day. For example, Filesystem with a single allowed directory:

claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem /path/to/dir

Use it when you need a narrow capability such as fetching web pages or a scoped folder outside your project. Claude Code already has native file tools for the project itself, so Filesystem matters more in Claude Desktop. Details: MCP reference servers and Filesystem MCP server.

How to choose: scopes and stacking

Every claude mcp add command takes a --scope flag. Picking the right one matters as much as picking the server:

Scope	Stored in	Loads in	Shared
local (default)	~/.claude.json, under the project	This project only	No
project	.mcp.json at the repo root	This project only	Yes, via git
user	~/.claude.json	All your projects	No

# Personal, everywhere: docs lookup
claude mcp add --scope user --transport http context7 https://mcp.context7.com/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

# Team, this repo: issue tracker everyone uses
claude mcp add --scope project --transport http linear-server https://mcp.linear.app/mcp

A practical split:

• User scope for servers you want in every project and that hold only your credentials, such as Context7.
• Project scope for servers the whole team should share, such as the issue tracker or error monitor for that repo. Never commit secrets in .mcp.json; reference environment variables with the ${VAR} syntax instead.
• Local scope for experiments and anything with personal tokens for one repo.

Suggested starter stacks:

You are…	Start with
Any developer	Context7 + GitHub
Frontend developer	+ Playwright MCP or Chrome DevTools MCP, + Figma
Backend developer	+ DBHub (read-only user), + Sentry
Team lead	+ Linear or Notion at project scope

Manage servers with claude mcp list, claude mcp get <name> and claude mcp remove <name>, and check status inside a session with /mcp.

Context cost

Claude Code enables tool search by default on supported models, so only tool names load at session start and full definitions load when Claude needs them. Adding a server is therefore cheap in context, but large tool outputs are not: Claude Code warns when an MCP tool result passes 10,000 tokens and caps it at 25,000 by default (raise it with MAX_MCP_OUTPUT_TOKENS). Ask for narrow queries, especially with DBHub and Sentry. See context management for /context and friends.

Security notes

The Claude Code docs are blunt about this: verify you trust each server before connecting it. Our rules of thumb:

1. Prefer first-party servers. Every pick above is maintained by the service owner or the MCP steering group.
2. Prefer OAuth over static tokens, and keep tokens in headers, never in URLs or committed files.
3. Least privilege. Fine-grained GitHub tokens, read-only database users, project-scoped servers only where the team needs them.
4. Watch for prompt injection. Servers that fetch external content (browsers, Fetch, issue trackers with public input) can carry instructions written by someone else. Combine them with sensible permission modes and hooks that block risky actions.
5. Remove what you don’t use. claude mcp remove <name> is one command.

MCP servers in your Claude kit

MCP servers are the “hands” of a Claude kit: they connect Claude to the systems where your work lives. Pair them with skills that tell Claude how to use each server in your workflow, and bundle the combination in a plugin so your team installs it in one step. If you want to build your own server, start with the mcp-builder skill and test it with MCP Inspector.