MCP Server Directories and Lists
MCP server lists. Finding MCP servers for AI agents on MCP directories, registries and marketplaces. Best MCP servers for Claude Code or Codex.
An MCP server is a small connector that gives an AI agent, such as Claude Code or Codex, an ability it doesn't have on its own. You'll find thousands of them listed online, in directories, registries, marketplaces, and lists on GitHub. Here I explain which places to look first, whether the listings can be trusted (most directories check very little), and how to judge a server before you add it. If MCP servers are new to you, start with MCP Servers on Mac.
I've got a list of servers I recommend for coding: see Best MCP Servers for Coding on Mac.
Before you get started
I recommend using our free app to add your first MCP servers. It is a FREE Mac app that installs the developer tools and AI coding agents you need, adds MCP servers to Claude Code and Codex, and verifies your setup is working. Here are all the details about the app before you download. Then download it:
MCP registries, directories and lists
You'll see four kinds of places that list MCP servers:
- App catalogs - lists built into an AI app, such as Claude's connectors directory. They are the smallest and the most carefully chosen, but also the most limited, and they lean heavily toward large commercial products rather than open source tools.
- Registries - a registry is a database of servers that other programs can read. The official MCP Registry is the main one. It records who publishes each server, but it doesn't evaluate them.
- Directories and marketplaces - websites with search, categories, and rankings. Some are run by companies that also sell hosting.
- Lists on GitHub - "awesome lists" are pages of links that anyone can add to, sorted by category. These are the least commercial and least "curated."
Without curation or filtering, you'll find an enormous number of MCP servers. When I looked, one large directory listed over 100,000 projects. The large directories collect every server anyone has published, including copies, experiments, and abandoned projects.
Start with your AI app's own catalog
The catalogs built into your AI app are the best place to start. Most of the servers are official versions for commercial products:
- Anthropic's connectors directory - remote MCP servers you connect from Claude with your own sign-in. Open it from
Settings > Connectorsin Claude Desktop, or browse it on the web at claude.com/connectors. Anthropic labels each connector. "Verified" means Anthropic tested its tools, though Anthropic says that isn't a security audit. "Community" means it passed automatic checks only. - Claude Code plugins - type
/pluginin Claude Code and choose Discover. Many plugins include MCP servers. Anthropic chooses the plugins but says it doesn't control the MCP servers inside them. - Codex plugins - type
/pluginsin Codex to browse the plugin catalog that Codex shares with ChatGPT.
The official MCP Registry
The official MCP Registry is run by the MCP project itself. It lists about 40,000 servers. To publish under a name, a publisher has to prove they own it, through their GitHub account or their web domain. That makes the registry the best place to confirm that a server really comes from the company it claims to come from.
The registry doesn't check servers for safety, and the MCP project says a listing isn't an endorsement. It is also still a preview, so its records can change. Use it to check a publisher, not to choose a server.
GitHub's MCP Registry
GitHub's MCP Registry is a short list of about 400 servers, with the publisher shown for each one. Each server's page shows its instructions from GitHub, and most of them give the setup steps for Claude Code and Codex. Because each server links to its code on GitHub, you can see how many people use it and how recently it was updated.
Docker's MCP Catalog
If you already use Docker Desktop, Docker's MCP Catalog offers over 300 local MCP servers packaged by Docker. Docker builds and signs each one, and each runs in its own container (a sealed-off space), so a server can't reach the rest of your Mac unless you allow it. That makes it a safer way to run local servers. It needs Docker Desktop 4.62 or later, which is more than most beginners want to install just for MCP servers.
Large MCP directories
The large directories list everything. They are useful for searching, but most check very little, and several show advertising or paid placements:
- Glama - about 98,000 servers. I'd look here first, because Glama grades each server for license, quality, and maintenance, and labels each tool "read-only" or "destructive" (able to change or delete things). Glama also sells hosting.
- Smithery - about 26,000 servers, with usage counts. Smithery can host servers and handle sign-ins for you, but connecting needs a Smithery account and some listings are paid.
- MCP Market - about 49,000 servers, with a "Top 100" leaderboard. It doesn't say who runs it.
- PulseMCP - about 22,000 servers, labeled by publisher type, with estimated weekly visitors. It has stopped taking new listings, so it is falling out of date.
- mcpservers.org - about 10,000 servers. It is the web version of one of the awesome lists, with paid sponsor slots.
- mcp.so and LobeHub - large community directories. LobeHub grades servers; mcp.so shows advertising and doesn't say what it checks.
Awesome lists
Awesome lists are long pages of links on GitHub, sorted by category. Anyone can suggest an addition, and no one checks the servers. They are good for browsing to see what exists:
- punkpeye/awesome-mcp-servers - the most popular list, with about 4,000 servers by category and icons that mark official, remote, and local servers. It links to Glama's grades.
- jaw9c/awesome-remote-mcp-servers - remote servers only. Its author argues that a remote server is easier to judge, because you can see whose web domain it runs on.
Zapier or Composio
Zapier and Composio are services that connect one MCP server to thousands of apps. Zapier MCP works with more than 9,000 apps, and each tool your agent uses counts against your Zapier plan. Composio offers over 1,600 toolkits. Both need an account and can cost money, and both see everything your agent sends through them.
Evaluate an MCP server
Before you add a server, check these things:
- Who publishes it - prefer the company that owns the product, such as Notion's own server for Notion. Confirm the publisher in the official MCP Registry or on the company's website. For companies that don't publish their own MCP servers for their products, you may need to use an MCP server written by someone in the community, and there may be several competing choices. Check GitHub stars and issues as you would any open source project.
- Remote or local - a remote server runs on the publisher's computers and nothing runs on your Mac. A local server runs on your Mac with your permissions, so it can read your files and run commands. The MCP project's security best practices warn that a local server can run any code.
- What it can do - read its list of tools. A server that only reads is safer than one that can change or delete things. Glama labels each tool for you.
- What it asks for - a server should ask for the fewest permissions it needs. Be wary of one that wants full access to an account when it only needs to read.
- Whether it is maintained - look at its GitHub page for recent updates and how the publisher responds to problems.
- What it reads - a server that reads outside content, such as web pages or email, can pass hidden instructions to your agent. This is called prompt injection, and Anthropic warns that a malicious MCP server may include hidden instructions of its own.
Anthropic's connector safety checklist and the OWASP MCP Top 10, a list of the most common MCP security risks, go into more detail.
How to add an MCP server to your AI agent
I've written a how-to article for Claude Code and Claude Desktop. See Add MCP Servers to Claude Code and Claude Desktop. For Codex, see Developer Setup: MCP Servers.
Continue setting up your Mac
Don't miss the full visual roadmap and checklist that shows how to set up a Mac for software development, with all the essential tools and settings you might not yet know about.