Best MCP servers: a curated list by job, with Claude Code install commands

What are the best MCP servers?

The best MCP servers are the official ones from the tools you already use: GitHub for code, Playwright MCP for browser testing, Context7 for current library docs, Supabase or DBHub for databases, Sentry for errors, Figma for designs, and Linear or Notion for project work. Each one lets an agent like Claude Code read and act in that system directly, instead of you pasting data into the chat.

This list includes only servers with an official repository or an official vendor docs page, each checked on 23 October 2026. Every install command comes from the vendor's docs or from Anthropic's Claude Code MCP docs. For the full mechanics of adding, scoping and securing servers, see Claude Code MCP.

How do you add an MCP server to Claude Code?

There are two shapes. A remote server runs at a URL:

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

A local server runs as a process on your machine, and everything after -- is its launch command:

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

Servers are saved at local scope by default (you, this project). Add --scope user to use a server in every project, or --scope project to write it to a shared .mcp.json. After adding, run claude mcp list or type /mcp inside a session to check the status and sign in to servers that use OAuth.

The list at a glance

Code and repositories

GitHub MCP Server

GitHub's official MCP server lets an agent search code, read and file issues, and work with pull requests. GitHub hosts a remote version at https://api.githubcopilot.com/mcp/ and also ships a local server as a Docker image or binary. The Claude Code docs add the remote server with a personal access token:

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

GitHub's own Claude install guide also gives claude mcp add-json and Docker variants.

Security: Use a fine-grained token limited to the repos the agent should touch. The local server supports --read-only, which skips every write tool, and --toolsets to load only the groups of tools you need. Also ask whether you need the server at all: Anthropic's Claude Code best practices call CLI tools like gh "the most context-efficient way to interact with external services."

Browser automation and debugging

Playwright MCP

Playwright MCP from Microsoft gives an agent a real browser it can navigate, click and read. It works from Playwright's accessibility tree rather than screenshots, so it does not need a vision model. It needs no account. The command from the Claude Code MCP quickstart:

claude mcp add playwright -- npx -y @playwright/mcp@latest

Security: The README states plainly that "Playwright MCP is not a security boundary." File access is limited to workspace roots by default, and one tool that runs arbitrary JavaScript is flagged as equivalent to remote code execution. The README also recommends the Playwright CLI with skills for coding agents, because CLI commands use fewer tokens; MCP fits long, stateful browsing sessions better.

Chrome DevTools MCP

Chrome DevTools MCP lets an agent control and inspect a live Chrome browser for automation, debugging and performance analysis. Its client configuration guide gives this command for Claude Code:

claude mcp add chrome-devtools --scope user npx chrome-devtools-mcp@latest

It can also be installed as a Claude Code plugin with bundled skills: run /plugin marketplace add ChromeDevTools/chrome-devtools-mcp, then /plugin install chrome-devtools-mcp@chrome-devtools-plugins.

Security: The README warns that it "exposes content of the browser instance to the MCP clients," so avoid using it in a browser profile signed in to sensitive accounts. Usage statistics are collected by default; turn them off with --no-usage-statistics.

Docs and search

Context7

Context7 by Upstash feeds an agent "up-to-date, version-specific documentation and code examples straight from the source," which helps with libraries that changed after the model was trained. Its all clients page gives the remote command for Claude Code:

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

The local version is claude mcp add --scope user context7 -- npx -y @upstash/context7-mcp --api-key YOUR_API_KEY. Keys come from the Context7 dashboard.

Security: Library docs are community-contributed and maintained by their owners, and Context7 says it "cannot guarantee the accuracy, completeness, or security of all library documentation." Treat retrieved docs as input, not instructions.

Claude Code docs server

Anthropic hosts a docs server with full text search over the Claude Code documentation. It needs no sign-in, which makes it the easiest server to test your setup with:

claude mcp add --transport http claude-code-docs https://code.claude.com/docs/mcp

Fetch (reference server)

The Fetch reference server turns web pages into text for a model. It starts with uvx mcp-server-fetch; to add a launch command like that, put it after -- as shown earlier. Claude Code already has built-in WebFetch and WebSearch tools (see permissions), so Fetch matters more in clients without them. Security: its README warns it "can access local/internal IP addresses and may represent a security risk."

Databases

Supabase MCP

Supabase's MCP server lets an agent "interact with and query your Supabase projects on your behalf": list tables, run SQL, read logs and manage branches. Supabase's docs give this command:

claude mcp add --scope project --transport http supabase "https://mcp.supabase.com/mcp?features=docs%2Caccount%2Cdatabase%2Cdebugging%2Cdevelopment%2Cfunctions%2Cbranching"

Security: Supabase calls prompt injection "the primary attack vector unique to LLMs." Its docs recommend scoping the server to one project, using read-only mode (SQL runs as a read-only Postgres user), keeping manual tool approval on, and connecting to production only "when the task requires production evidence."

DBHub

DBHub from Bytebase is a minimal server for PostgreSQL, MySQL, MariaDB, SQL Server, Oracle and SQLite, with two default tools: run SQL and search schema objects. The Claude Code docs use it as their database example:

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

Security: Put a read-only database user in the connection string, as the example does, so a bad query can't change data.

Errors and monitoring

Sentry MCP

Sentry's MCP server lets an agent search errors, triage issues and analyze performance, and it uses OAuth for every connection. From the Claude Code docs:

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

Security: Sentry's own setup page scopes the URL to one project, https://mcp.sentry.dev/mcp/{organizationSlug}/{projectSlug}, and says "Scoping to a project is recommended when possible." Scoped connections hide discovery tools you don't need.

Design

Figma MCP

The Figma MCP server gives agents design context from your files for generating code, and can write content back to the canvas. Figma's remote server installation page recommends its Claude Code plugin, claude plugin install figma@claude-plugins-official, or this manual command:

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

Plans and security: Figma lists the server on Starter, Professional, Education, Organization, Government and Enterprise plans; the desktop server needs a Dev or Full seat on a Professional, Education, Organization, Government or Enterprise plan. Writing to the canvas outside your drafts needs a Full seat.

Project and knowledge tools

Linear MCP

Linear's MCP server offers tools for "finding, creating, and updating objects in Linear like issues, projects, and comments." It uses OAuth; finish sign-in with /mcp after adding:

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

Security: The default endpoint can write. For read-only access, Linear offers https://mcp.linear.app/mcp/readonly.

Notion MCP

Notion MCP is "a remote MCP server hosted by Notion" that can search, read, create and update pages and databases your account can access. From Notion's getting started page:

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

Security: The agent sees everything your Notion account can see. Workspace owners can manage MCP client access in Notion's settings.

Stripe MCP

Stripe's MCP server gives agents tools to read and write through the Stripe API and to search Stripe's docs. Stripe's docs give:

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

Security: Stripe requires human confirmation before some write actions, such as refunds, and asks you to keep your client's tool approval on and to be careful combining it with other servers because of prompt injection. Stripe also says that from December 2026 its MCP server will stop accepting full-access secret keys and restricted keys without the Agent tag; use OAuth or an agent API key.

The official reference servers (and what was archived)

The modelcontextprotocol/servers repo now holds a small set of reference servers maintained by the MCP steering group: Everything, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time. The repo says they are "intended as reference implementations to demonstrate MCP features and SDK usage," not production-ready products.

Many servers that older lists still recommend have been moved to servers-archived: AWS KB Retrieval, Brave Search, EverArt, GitHub, GitLab, Google Drive, Google Maps, PostgreSQL, Puppeteer, Redis, Sentry, Slack and SQLite. Brave Search has been replaced by Brave's own official server, Slack is now maintained by Zencoder, and GitHub and Sentry have the vendor servers listed above. If a guide tells you to install @modelcontextprotocol/server-github or server-postgres, it is out of date.

The Filesystem server is useful in chat apps that can't touch your files. Claude Code already reads and edits files in your project, so you rarely need it there.

Where to find more MCP servers

• The Anthropic Directory. The Claude Code docs describe it as "reviewed connectors," and any remote server listed there can be added with claude mcp add.
• The official MCP Registry. The official metadata catalog of publicly accessible servers. Its repo says it is still a preview release. It verifies who owns a server's namespace, but it does not say it reviews servers for safety, so vet them yourself.

How to choose an MCP server

1. Start from friction. Add a server when you notice yourself copying data from a tool into the chat. One or two well-chosen servers beat ten idle ones.
2. Prefer the vendor's own server. The company that runs the service is best placed to keep its server in step with the API. Community servers can be fine, but read the code first.
3. Prefer remote with OAuth for SaaS tools. You sign in once in the browser, with no long-lived token on disk.
4. Check for a CLI first. If a good CLI exists, such as gh for GitHub or the Stripe CLI, Claude Code may do the job with less setup and less context.
5. Use the narrowest access. Read-only endpoints, project scoping and read-only database users limit the damage of a wrong call or an injected instruction.
6. Keep personal tokens out of git. Use local or user scope for servers with your credentials, and reference secrets with ${VAR} in a shared .mcp.json.

Claude Code's docs put the core rule in one line: "Verify you trust each server before connecting it. Servers that fetch external content can expose you to prompt injection risk."

MCP servers vs skills

An MCP server gives an agent access to a system. A skill gives it a method: instructions, and sometimes scripts, for a task such as triaging Sentry issues the way your team does. They work best together, and several vendors above (Chrome DevTools, Figma, Stripe) now ship skills next to their servers for that reason. The skills vs MCP guide covers when to pick which, and Claude Code plugins bundle both into one install.

Your servers are configured per agent, but the skills that explain how to use them are worth keeping in one place. Kipt stores your prompts and skills as Markdown in a private GitHub repo you own and installs every skill into Claude Code or Codex with one npx skills add command. If you have a prompt you use with a server, such as your bug triage steps, the free SKILL.md generator turns it into a skill. For skills worth installing first, see the best Claude Code skills.