What is MCP?

MCP — the Model Context Protocol — is how an AI assistant stops guessing and starts doing. It's the difference between Claude describing what a Stripe refund would look like and Claude actually issuing one.

The problem it solves

A language model on its own is sealed in a box. It can reason about your database, but it can't query it. Before MCP, every combination of AI app and external tool needed its own bespoke integration: if there were 10 assistants and 100 tools, someone had to write 1,000 connectors.

MCP replaces that with one standard. A vendor builds a single server for their product, and every compatible client can use it — the same reason USB replaced a drawer full of proprietary cables.

What an MCP server actually is

A server is a small program that advertises a set of tools — named actions with typed inputs, like create_issue or run_query. Your AI client reads that list, decides when a tool would help, calls it, and feeds the result back into the conversation. You approve the calls that matter.

Servers come in two shapes, and the difference decides how you install them:

Remote

Hosted by the vendor

Lives at a URL like mcp.linear.app. Nothing to install — you add the URL to your client and sign in with OAuth. The vendor runs, updates and secures it.

These are invisible to GitHub-scraping directories, which is why so many "no official server" claims elsewhere are wrong.

Local

Runs on your machine

Ships as an npm or PyPI package. Your client launches it as a subprocess and talks over stdio. You supply the API keys, and they stay on your machine.

Local servers can reach your filesystem and network, so the code you run matters.

Official vs community — why we split them

Anyone can write an MCP server for anyone's product. That's the ecosystem's strength and its sharpest edge. An official server is maintained by the company behind the product: it follows their API changes, and it has been through whatever security review they apply to their own code. A community server is someone else's side project wearing the brand's name.

Plenty of community servers are excellent and predate the official ones. But they are unaudited third-party code that you hand real credentials to. Independent research has found roughly a third of public MCP servers carry SSRF vulnerabilities, and 41% ship with no authentication at all. Before connecting one: read the source, check the maintainer, look at when it was last committed to, and issue it a scoped, revocable key rather than an account-wide token.

That distinction is the whole reason this site exists. Every verdict here says which of the two you're looking at, and shows the evidence. Here's how we decide →

Getting started

  1. Find out whether the tool you want even has a server. That's the search box on the front page — 234 brands, 212 with official servers.
  2. Pick your client. Setup differs per app: Claude, Cursor, ChatGPT. Each brand page here has copy-paste config for all of them.
  3. Start with something read-only. A docs or search server is a low-stakes way to feel how tool calling behaves before you give an assistant write access to production.

Common questions

What does MCP stand for?

Model Context Protocol. It's an open standard for connecting AI assistants to external tools and data sources, introduced by Anthropic in late 2024 and since adopted across the industry.

What is an MCP server?

A small program that exposes one product's capabilities — reading a database, creating an issue, searching documents — in a format any MCP-compatible AI client can call. One server per product; any client can use it.

Is MCP only for Claude?

No. MCP started at Anthropic but is an open protocol. Claude, ChatGPT, Cursor, VS Code, Zed, Windsurf and others can all connect to MCP servers, which is the point: one server works everywhere.

What's the difference between an official and a community MCP server?

An official server is built and maintained by the company whose product it connects to, so it tracks their API and carries their security review. A community server is third-party code doing the same job unofficially — often excellent, sometimes abandoned, always unaudited by the vendor.

Do I need to install anything to use an MCP server?

It depends on the server. Remote servers are hosted by the vendor and need only a URL plus sign-in. Local servers run on your own machine, usually launched with npx or uvx by your AI client.

Ready to check a specific tool?

Hit ⌘K anywhere on this site, or start from the full list.

Browse all 234 verdicts →