Guide · MCP servers
Claude MCP servers — the ones worth installing
MCP servers give Claude real tools: your files, your repos, a browser, your database, your error tracker. Here’s what they are, how they relate to claude.ai connectors, and a short list of the servers most people actually use.
Educational guide · updated June 2026.
# add a local (stdio) server
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/projects
# add a remote (HTTP) server
claude mcp add --transport http pagelive https://app.pagelive.io/api/mcp
# inside Claude Code: list servers + finish OAuth
/mcpThe short answer
What are MCP servers?
MCP servers are programs that give Claude tools and data through the Model Context Protocol — an open standard for connecting AI assistants to outside systems. In Claude Code you add one with a single claude mcp add command; on claude.ai, remote MCP servers show up as connectors you enable in Settings.
One standard, two doors
MCP servers vs claude.ai connectors
Same protocol, different packaging. Connectors are the consumer-friendly wrapper on claude.ai; in Claude Code you manage servers directly — including local ones that never leave your machine.
| claude.ai connectors | MCP servers in Claude Code | |
|---|---|---|
| Where you use it | claude.ai (web, desktop, mobile) | Claude Code (terminal) and other MCP clients |
| How you add it | Settings → Connectors → enable or paste a URL | claude mcp add … or a shared .mcp.json |
| What runs | Remote, hosted MCP servers | Local (stdio) or remote (HTTP) servers |
| Auth | OAuth sign-in in the browser | OAuth via /mcp, env vars, or API keys |
| Protocol underneath | MCP — the same standard | MCP — the same standard |
Not a developer? The connector side has its own guide: Claude connectors — the list.
The list
Widely-used MCP servers for Claude
A short list of servers with real adoption — one factual sentence each, no rankings. Pagelive is ours; the rest we simply use or see used everywhere.
Filesystem
Read and write files in directories you allow — the reference server for local file access.
GitHub
Search repos, read code and issues, and manage pull requests from the conversation.
Google Drive
Search and read your Drive files so Claude can work from real documents.
Slack
Search messages and channels in your Slack workspace and draft updates with that context.
Postgres
Run read-only SQL queries against your database and inspect schemas.
Playwright / Puppeteer
Drive a real browser — navigate, click, fill forms, and take screenshots.
Sentry
Pull issues, errors, and stack traces from your Sentry projects for debugging.
Stripe
Search Stripe’s documentation and work with customers, products, and payments via the API.
Cloudflare
Manage Workers, DNS records, and account resources, and query docs and analytics.
Notion
Search, read, and update pages and databases in your Notion workspace.
Pagelive
Publish the HTML Claude builds as a live, tracked, noindex-by-default link — and update it at the same URL.
See the connector →Install pattern
How to add a server to Claude Code
Local servers run as a command on your machine; remote servers are just a URL. Either way it’s one line, then /mcp to verify and authenticate.
# add a local (stdio) server
claude mcp add filesystem -- npx -y @modelcontextprotocol/server-filesystem ~/projects
# add a remote (HTTP) server
claude mcp add --transport http pagelive https://app.pagelive.io/api/mcp
# inside Claude Code: list servers + finish OAuth
/mcpServers added with --scope project land in a shared .mcp.json, so the whole team gets them on checkout.
Publish & share what you build
The Pagelive MCP server, tool by tool
This one is ours, so we’ll keep it factual: seven tools that turn Claude’s HTML output into live, tracked, noindex-by-default links — and manage them afterwards.
publish_page Publish HTML as a live, tracked link — noindex by default, with an optional password.
update_page Re-publish edits to the same URL — targeted {old_str, new_str} edits or a full rewrite.
list_pages List your published pages with links and view counts.
get_page_stats Opens, unique viewers, country, referrer, and time on page for one page.
list_form_submissions Read the replies a page collected through a plain reply form.
list_domains List the custom domains connected to your workspace.
list_workspaces List the workspaces you can publish into (personal + teams).
It’s a remote server: https://app.pagelive.io/api/mcp — add it to Claude Code with the command above, or one-click connect on claude.ai via Settings → Connectors. Setup details in the MCP quickstart; the bigger picture is on the connector page and publish from Claude.
Free for your first 10 pages; Pro is $15/mo ($150/yr) for custom domains and more — see pricing.
Claude MCP servers — FAQ
What is an MCP server? +
An MCP server is a program that gives Claude tools and data through the Model Context Protocol — an open standard for connecting AI assistants to outside systems like files, databases, browsers, and APIs. Claude calls the server’s tools during a conversation; the server does the work and returns the result.
What’s the difference between MCP servers and Claude connectors? +
They’re the same technology. A connector is claude.ai’s packaging of a remote MCP server — you enable it in Settings → Connectors and sign in with OAuth. In Claude Code you add servers yourself with claude mcp add, and you can also run local servers that work with your filesystem or database directly.
How do I add an MCP server to Claude Code? +
Run claude mcp add <name> -- <command> for a local server, or claude mcp add --transport http <name> <url> for a remote one. Then run /mcp inside Claude Code to check status and complete OAuth where needed. Project-level servers can be checked into .mcp.json so your team shares them.
What are the best MCP servers for Claude Code? +
It depends on the job. For most developers the staples are filesystem and GitHub access, a browser-automation server like Playwright or Puppeteer for testing, Sentry for debugging production errors, and a database server like Postgres for read-only queries. If you publish what you build, Pagelive turns Claude’s HTML output into a live, tracked link.
Are MCP servers safe to use? +
An MCP server can do whatever its tools allow with the access you give it, so treat servers like any dependency: install only from developers you trust, prefer official servers, and scope credentials narrowly — read-only database users, least-privilege API keys. Local servers run on your machine; remote servers authenticate with OAuth and can be disconnected at any time.
Prefer the claude.ai side? Start with Claude connectors. For Pagelive specifically: the connector · quickstart.
Give Claude a publish tool: ship what it builds as a secure link.
One claude mcp add, then just ask Claude to publish. Free to start · noindex by default.