The best MCP servers for Claude Code in 2026 are GitHub, Context7, Playwright, Chrome DevTools and Sentry, followed by a database server and the connector for whatever tracker your team uses. Most of them are now hosted by the vendor, so adding one is a single claude mcp add --transport http command with no local install. Several servers that older lists still recommend, including the original GitHub, Postgres, Puppeteer and Slack reference servers, are archived and no longer maintained. This list uses the current, vendor-maintained versions, checked against the Claude Code MCP docs and each vendor’s own instructions on October 19, 2026.
What an MCP Server Does for Claude Code
The Model Context Protocol is an open standard for connecting an AI client to outside tools and data. An MCP server exposes a set of tools, such as “list open pull requests” or “run this query”, and Claude Code calls them as part of its work. Claude Code can already read files, edit code and run shell commands on its own, so the servers worth adding are the ones that reach things it cannot: your repository host, current library docs, a real browser, production errors, your database and your team’s tracker.

How to Add an MCP Server to Claude Code
There are two kinds of server. A remote server runs at the vendor and you connect over HTTP. A local server is a program Claude Code starts on your machine (the stdio transport). The commands:
# Remote server (recommended where the vendor offers one)
claude mcp add --transport http <name> <url>
# Local server: everything after -- is the command Claude Code runs
claude mcp add --transport stdio <name> -- npx -y <package>
# Manage them
claude mcp list
claude mcp get <name>
claude mcp remove <name>The SSE transport still works but is deprecated, so use HTTP for anything new. For servers that sign in with OAuth, run /mcp inside a session and follow the browser login. Pass an API key with --header "Authorization: Bearer ..." for remote servers or --env KEY=value for local ones.
| Scope | Stored in | Who gets the server |
|---|---|---|
| Local (default) | ~/.claude.json | Only you, only in this project |
| Project | .mcp.json in the repo root | Everyone who clones the repo, after they approve it |
| User | ~/.claude.json | Only you, in every project |
Add --scope project or --scope user to choose. Project servers are shared through git, and Claude Code asks each person to approve them before they run. Inside .mcp.json, write ${VAR} or ${VAR:-default} to read secrets from environment variables instead of committing them.

The 10 Best MCP Servers for Claude Code
1. GitHub
Maintained by GitHub. Transport: Remote (HTTP) or Docker. Issues, pull requests, code search across repos and repository workflows. The highest-impact single install for most developers. Use the remote server and a personal access token, or run GitHub’s container locally.
claude mcp add --transport http github https://api.githubcopilot.com/mcp \
-H "Authorization: Bearer YOUR_GITHUB_PAT"2. Context7
Maintained by Upstash. Transport: Remote (HTTP). Current, version-specific documentation for the libraries in your project, so Claude stops writing against an API that has since changed. The guided setup handles sign-in and the key for you.
npx ctx7 setup --claude
# or manually
claude mcp add --transport http context7 https://mcp.context7.com/mcp \
-H "Authorization: Bearer YOUR_CONTEXT7_KEY"3. Playwright
Maintained by Microsoft. Transport: Local (stdio). A real browser Claude can navigate, click and read, for end-to-end tests and for checking a UI change it just made. No account needed, which makes it the easiest first server to try.
claude mcp add playwright -- npx @playwright/mcp@latest4. Chrome DevTools
Maintained by Google. Transport: Local (stdio). Connects to a live Chrome instance to read console errors, inspect network requests and record performance traces. Pick this over Playwright when the job is debugging rather than testing.
claude mcp add chrome-devtools --scope user -- npx chrome-devtools-mcp@latest5. Sentry
Maintained by Sentry. Transport: Remote (HTTP). Production errors, stack traces and issue context inside the session, so Claude can work from a real incident to a fix. Sign in with /mcp after adding it.
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp
6. DBHub (Postgres, MySQL and others)
Maintained by Bytebase. Transport: Local (stdio). Gives Claude your schema and lets it run queries, which is the example Anthropic uses in its own docs. Connect it with a read-only database user, never an admin account.
claude mcp add --transport stdio db -- npx -y @bytebase/dbhub \
--dsn "postgresql://readonly:pass@host:5432/db"7. Supabase
Maintained by Supabase. Transport: Remote (HTTP). Database, schema, docs and project management for Supabase projects. Scope it to one project and turn on read-only mode in the URL until you trust the workflow.
claude mcp add --transport http supabase \
"https://mcp.supabase.com/mcp?project_ref=YOUR_REF&read_only=true"8. Linear
Maintained by Linear. Transport: Remote (HTTP). Read and update issues, projects and cycles, so a ticket can go from description to branch to pull request without leaving the terminal.
claude mcp add --transport http linear https://mcp.linear.app/mcp9. Atlassian (Jira and Confluence)
Maintained by Atlassian. Transport: Remote (HTTP). Jira issues and Confluence pages for teams on Atlassian. Use the v2 endpoint, which Atlassian marks as the recommended version, and authenticate with /mcp.
claude mcp add --transport http atlassian https://mcp.atlassian.com/v2/mcp10. Figma
Maintained by Figma. Transport: Remote (HTTP). Design context from Figma files, so Claude builds components from the actual frames, spacing and tokens instead of a screenshot description.
claude mcp add --transport http figma https://mcp.figma.com/mcpQuick Comparison
| Server | Best for | Transport | Needs an account |
|---|---|---|---|
| GitHub | Repos, issues, pull requests | HTTP or Docker | Yes (token) |
| Context7 | Current library docs | HTTP | Yes (API key) |
| Playwright | End-to-end testing | stdio | No |
| Chrome DevTools | Browser debugging | stdio | No |
| Sentry | Production errors | HTTP | Yes (OAuth) |
| DBHub | SQL databases | stdio | Your database user |
| Supabase | Supabase projects | HTTP | Yes |
| Linear | Issue tracking | HTTP | Yes (OAuth) |
| Atlassian | Jira and Confluence | HTTP | Yes (OAuth) |
| Figma | Design to code | HTTP | Yes |
Also Worth Knowing
Notion (https://mcp.notion.com/mcp), Stripe (https://mcp.stripe.com) and Cloudflare (https://mcp.cloudflare.com/mcp) all run official remote servers you add the same way. For web search, Brave now maintains its own server, @brave/brave-search-mcp-server. The small set of reference servers maintained by the MCP steering group, Fetch, Filesystem, Git, Memory, Sequential Thinking and Time, is still current, but Filesystem and Git rarely add much in Claude Code, which already reads files and runs git itself.
Archived Servers to Stop Installing
The modelcontextprotocol/servers repository now keeps only the steering-group reference servers. The rest moved to an archive that is marked as no longer maintained. If a tutorial tells you to install one of these, use the replacement instead.
| Archived server | Use instead |
|---|---|
GitHub (@modelcontextprotocol/server-github) | GitHub’s official server, remote or Docker |
| PostgreSQL and SQLite | DBHub, Supabase, or your database vendor’s own server |
| Puppeteer | Playwright (Microsoft) or Chrome DevTools (Google) |
| Sentry (reference version) | Sentry’s hosted server at mcp.sentry.dev |
| Slack | The Slack server now maintained by Zencoder |
| Brave Search (reference version) | Brave’s official @brave/brave-search-mcp-server |
| Google Drive, GitLab, Redis, Google Maps | A vendor-maintained server from the MCP Registry |
Find more servers
The official MCP Registry is the place to look for anything not on this list. Prefer servers published by the company that owns the tool.
How Many MCP Servers Can You Run at Once?
More than older advice suggests. Claude Code now uses tool search by default: at the start of a session it loads only each server’s tool names and instructions, and fetches the full tool definitions when a tool is needed, so adding servers has little effect on your context window. The real limits are elsewhere. Output from a single tool call is capped at 25,000 tokens by default (raise it with MAX_MCP_OUTPUT_TOKENS), slow servers can be given more startup time with MCP_TIMEOUT, and every server you connect is another thing that can fail. Keep the ones you use, and remove the rest with claude mcp remove.

If a server hangs or a session stalls, check whether the problem is the server or the service before you change anything: check Claude’s status first, then run /mcp to see which servers are connected. More fixes are in the troubleshooting library.
Security: Treat Every Server as Code You Run
Anthropic’s docs say to verify you trust each server before connecting it, and warn that servers which fetch outside content can expose you to prompt injection. In practice: prefer vendor-hosted servers, give database servers read-only credentials, keep tokens in environment variables rather than .mcp.json, and review a project’s .mcp.json before you approve it, because approving it lets those programs start on your machine.

Servers, Plugins and Where to Go Next
An MCP server is one connection. A plugin can package that connection with the skills, agents and hooks that tell Claude when to use it, and several servers on this list ship as plugins too. The best Claude Code plugins list covers those, the Claude MCP servers guide explains the protocol in more depth, and Claude Code plan mode is how to review what Claude intends to do with a new server before it does it. Everything else is in the guides library.
For a company that wants the same servers, credentials policy and review rules on every developer’s machine, Strategized.ai builds private Claude plugin libraries that package them, with configuration included and any custom code scoped separately.
Give Claude your business in twenty minutes
The free Foundation Plugin from Strategized.ai interviews you and writes the Business Blueprint every Claude conversation starts from. No account, no card.
Get the Foundation Plugin →Free. Sent to your email.
Frequently Asked Questions
What are the best MCP servers for Claude Code?
For most developers: GitHub, Context7, Playwright, Chrome DevTools and Sentry, plus a database server such as DBHub or Supabase and the connector for your tracker (Linear or Atlassian). GitHub, Context7, Sentry, Linear, Atlassian, Supabase and Figma all run as hosted servers you add with one command.
How do I add an MCP server to Claude Code?
Run claude mcp add –transport http name url for a hosted server, or claude mcp add name — command for a local one. Add –scope project to share it through .mcp.json, and run /mcp inside a session to sign in to servers that use OAuth.
Where does Claude Code store MCP server settings?
Local and user servers are stored in ~/.claude.json. Project servers go in .mcp.json at the repository root, which is committed to git, and Claude Code asks each person to approve them before they run.
Is the official GitHub MCP server the same as @modelcontextprotocol/server-github?
No. The old reference server is archived and no longer maintained. GitHub now maintains its own server, available as a hosted endpoint at api.githubcopilot.com/mcp or as a Docker image.
How many MCP servers can I use in Claude Code at once?
There is no fixed limit, and tool search, which is on by default, loads only tool names at the start of a session, so extra servers cost little context. Keep the list to servers you actually use, because each one is another dependency that can time out.
Are MCP servers safe to use with Claude Code?
Only trusted ones. Anthropic warns that servers fetching outside content can expose you to prompt injection. Prefer servers maintained by the vendor of the tool, use read-only credentials for databases, and review a project’s .mcp.json before approving it.



