Best MCP Servers for Claude Code: 10 We'd Set Up First

The MCP servers worth setting up first in Claude Code for client work are GitHub, Playwright, Supabase or a Postgres connector, Context7, Sentry, Vercel, Stripe, Notion and HubSpot. Each connects Claude to a tool you'd otherwise copy and paste from. Add them one at a time, read-only where possible.

"Best" depends on your stack, so treat this as the list we'd start with for building client tools, and why. Every install command below was checked against the Claude Code MCP docs or the server's own docs in October 2026. Server names and endpoints change, so check the linked page before you install.

How MCP works in Claude Code (one minute)

MCP (Model Context Protocol) is an open standard for connecting AI tools to outside systems. An MCP server exposes tools, like "list open issues" or "run this SQL query," that Claude can call.

Add a remote server with:

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

Servers that use OAuth are authenticated inside Claude Code by typing /mcp and following the browser login. Scope controls who gets the server:

• local (default): you, in this project only
• project (--scope project): saved in .mcp.json at the project root, shared with everyone who clones the repo
• user (--scope user): you, in every project

Check a server is connected with /mcp or claude mcp list.

1. GitHub

What it does: reads and manages repos, issues, pull requests and code search.

Use it for: reviewing a client repo, triaging issues, opening pull requests and checking CI results without leaving the session.

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

Use a fine-grained personal access token limited to the repos you need. The server also supports a read-only mode and toolsets that limit what it can do (github/github-mcp-server).

2. Playwright

What it does: drives a real browser: open pages, click, fill forms, take snapshots.

Use it for: testing the web app you just built the way a user would, checking a client's booking flow or reproducing a bug.

claude mcp add playwright npx @playwright/mcp@latest

Command from the microsoft/playwright-mcp repo. It runs locally, so it uses your machine's browser and network.

3. Supabase

What it does: works with a Supabase project: database queries, schema, logs and more.

Use it for: designing tables, debugging a query, checking data while you build.

claude mcp add --scope project --transport http supabase \
  "https://mcp.supabase.com/mcp?project_ref=YOUR_PROJECT_REF&read_only=true"

Then run /mcp and authenticate. project_ref limits it to one project and read_only=true runs queries as a read-only Postgres user. Supabase recommends connecting to production only when you need production data, and says not to give the server to your customers, because it runs with your developer permissions (Supabase MCP docs).

4. Postgres (any database, via DBHub)

What it does: connects Claude to a Postgres database that isn't on Supabase.

Use it for: client projects on their own database.

The Claude Code docs use DBHub as their Postgres example. In a project's .mcp.json, with the connection string kept in an environment variable rather than in the file:

{
  "mcpServers": {
    "db": {
      "command": "npx",
      "args": ["-y", "@bytebase/dbhub", "--dsn", "${DATABASE_URL_READONLY}"]
    }
  }
}

Point it at a read-only database user. Never commit real credentials to .mcp.json.

5. Context7

What it does: pulls current documentation for libraries and frameworks into the session.

Use it for: avoiding outdated answers about fast-moving tools, which matters when a library changed last month.

Context7's docs give a setup command that targets Claude Code:

npx ctx7 setup --claude

You can also install it as a Claude Code plugin with /plugin marketplace add upstash/context7 and then /plugin install context7@context7-marketplace (Context7 docs). A free API key gives higher rate limits.

6. Sentry

What it does: reads errors and issues from Sentry.

Use it for: asking Claude to look at the latest production error in a client app and propose a fix.

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

Then run /mcp to sign in. No token goes in the command.

7. Vercel

What it does: searches Vercel docs, manages projects and deployments, reads deployment logs and queries Web Analytics.

Use it for: finding out why a deploy failed and checking a client site after launch.

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

Then /mcp to sign in. Vercel notes that the connection gets the same access as your Vercel account, and recommends keeping human confirmation on for changes (Vercel MCP docs).

8. Stripe

What it does: works with Stripe data and the Stripe API.

Use it for: building payment flows and checking products, prices or test transactions while you build.

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

Use test mode while building, and be careful connecting it to a live account.

9. Notion

What it does: reads and updates Notion pages and databases.

Use it for: pulling a client's specs, SOPs or meeting notes into a build, or writing documentation back.

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

10. HubSpot

What it does: works with HubSpot CRM data.

Use it for: client projects that read contacts and deals or build reports on top of the CRM.

claude mcp add --transport http hubspot --scope user https://mcp.hubspot.com/anthropic

That command is from the Claude Code docs' scope examples.

Security checklist

MCP servers give Claude real access to real systems. Before you add one for client work:

• Trust the source. Use official servers from the vendor or well-known repos. Stdio servers run on your machine with your permissions.
• Start read-only. Use read-only modes, read-only database users and narrow tokens. Add write access only when a task needs it.
• Scope tokens tightly. One token per client, limited to the repos, projects or data you need, so you can revoke it cleanly.
• Keep secrets out of .mcp.json. Use ${VAR} references and environment variables. The file is often committed.
• Watch for prompt injection. Servers that fetch outside content (web pages, emails, issues) can pass along text that tries to instruct Claude. Keep approvals on for actions that change things.
• Review project servers. Claude Code asks before using servers from a project's .mcp.json in interactive sessions. Read the file before approving, especially in repos you didn't write.

Learn to build client tools with it

The Stride AI Coding course teaches Claude Code and Codex for building custom tools for clients, including connecting them to the systems clients already use, with weekly live calls. Many of the free guides walk through specific MCP server setups.

FAQ

What is an MCP server in Claude Code?

It's a connector that gives Claude Code tools for an outside system, such as GitHub, a database or a payment platform. You add it with claude mcp add, and Claude can then call its tools during a session, with your approval where required.

How do I add an MCP server to Claude Code?

For a remote server, run claude mcp add --transport http <name> <url>, then type /mcp in Claude Code to authenticate if it uses OAuth. Use --scope project to save it in .mcp.json for your team, or --scope user for all your projects.

Are MCP servers safe?

They're as safe as the server and the access you give it. Use official servers, start read-only, scope tokens to what you need, keep secrets out of committed files and keep approvals on for actions that change data.

How many MCP servers should I install?

Only the ones a project needs. Each server adds tools Claude has to choose between, and each one is access you have to manage. Start with two or three for the current build and add more when a task calls for them.