Claude Code MCP — the setup guide that doesn't waste your time

Claude Code is the fastest way to use MCP servers. Install a server, register it once, and the agent knows about it for every session in that scope. Here's the exact sequence — and the servers worth installing first.

The 60-second install

Three commands. That's the whole flow.

# Install the GitHub MCP server, scoped to the current project
claude mcp add github -- npx -y @modelcontextprotocol/server-github

# Pass an env var the server needs
claude mcp add github --env GITHUB_PERSONAL_ACCESS_TOKEN=ghp_xxx -- npx -y @modelcontextprotocol/server-github

# Verify
claude mcp list

Now start a session with claude and ask "list my open PRs in raneyraines/foo". The agent will discover the GitHub tools and call them.

Scope: project vs user vs system

Claude Code supports three scopes for MCP servers. Choose deliberately.

--scope project (default) — server is registered for the current directory and committed to a project config file. Good for team setups where everyone gets the same tools.
--scope user — server lives in your user-level config. Available across every project. Good for personal tools (your filesystem server, your Linear server).
--scope local — registered to the current directory only, not committed. Good for experimentation.

The five MCP servers I install on every project

Filesystem — read and write local files with configurable access controls. claude mcp add fs -- npx -y @modelcontextprotocol/server-filesystem /path/to/project
Git — search, diff, and commit. claude mcp add git -- uvx mcp-server-git --repository .
GitHub — issues, PRs, code search, repos. Requires a personal access token.
Fetch — pull a URL into context as cleaned-up Markdown. Free, no key.
Memory — a persistent knowledge graph the agent can read and write across sessions.

Working with credentials

Most useful MCP servers need an API key. Three ways to provide them, in order of safety:

Per-server env flags — claude mcp add stripe --env STRIPE_API_KEY=sk_test_xxx -- npx -y @stripe/mcp-server. Stored in Claude Code's config.
Shell environment — export the variable in your shell. The server inherits it. Doesn't leak into a config file.
OS keychain / 1Password CLI — wrap the install command in op run -- or a keychain helper. Best for shared machines.

Never paste a production key into a project-scope server that's committed to git. That's how secrets end up on the internet.

Debugging MCP in Claude Code

If a server isn't showing up, walk this list:

claude mcp list — is it registered? Right scope?
claude mcp logs <server> — most server crashes are visible here.
Try the install command standalone in a terminal. If it fails outside Claude Code, the issue is upstream.
Check the server's README for required env vars and supported transports.

Pick your next server

The ClawHire directory has 2,500+ servers across 50+ categories. Browse by what you actually use day-to-day:

Databases — Postgres, MySQL, MongoDB, Redis, ClickHouse, BigQuery, more.
Developer tools — CI/CD, package managers, linters, build systems.
Browser automation — Playwright, Puppeteer, headless Chrome.
Knowledge & Memory — vector stores, knowledge graphs, persistent notes.
Version control — GitHub, GitLab, Bitbucket, Git itself.

FAQ

Does Claude Code support remote/SSE MCP servers?

Yes — pass --transport sse or --transport http with a URL instead of a command. The server runs as a remote service; Claude Code talks to it over HTTPS.

Can I use the same MCP server in Cursor and Claude Code?

Yes. The server doesn't care which client connects. Both clients have their own registration command, but the server config is identical.

What if a server I want isn't in the directory?

Browse the marketplace or check wong2's awesome list directly. New servers get indexed on the next catalog refresh.