New to MCP? Two minutes and you're running.
Your AI assistant (Claude, Cursor, and the like) can plug into outside tools: your database, GitHub, web search, thousands more. The plug is a standard called MCP. SocketCat is where you find those tools and add them, without guessing which are any good.
Open a server, pick your client (Claude, Cursor, VS Code, Cline), and hit copy. That is the whole setup.
Drop it into your client's config, restart, and your assistant can use the tool. No accounts, no build step.
Not sure which client you have? The config tabs on every server page cover the common ones.
How to read a server page
Every server shows the same few signals in the same spots, so once you learn them here you can size up any tool in a couple of seconds.
A clear A-to-E score built from three things anyone can check: is it maintained, is it complete, and does it work. Popularity is kept out on purpose. Open the panel on a server to see the exact math.
Who published it, and what the official registry proved. Publisher means a company's own verified domain. Verified means a proven GitHub identity. No badge means community, so check the source first.
Flags a known security advisory in the tool's dependencies. Green means none known, not a guarantee, so still read what it can access.
Exactly what the tool can touch before you run it: your secrets, the network, your files. Read this first for anything that handles sensitive data.
For hosted tools you can list and call the real tools right in the page. For local ones you get a one-line command to test it on your machine.
Three ways to find the right tool
The servers page: every listing graded A to E, with filters for category, quality, trust, and hosting.
Know what the tool should do? Tool search searches the tool schemas themselves. Try “take a screenshot” or “query postgres”.
Torn between two? Open either server and hit Compare for a side-by-side: grades, a diff of their tools, and a verdict.
Or press ⌘K anywhere to search servers, tools, and blueprints at once.
And when one tool isn't enough: blueprints
A blueprint is a ready-made team of AI agents that work together (one researches, one writes, one checks). Open one and you can read exactly how it works and export runnable code for LangGraph or CrewAI, wired to real tools. There are 130 to fork.
Found a combination that works? Share it as a stack
Save servers to your toolbox as you browse (the ☆ on any card), then turn it into a stack: a named page under your handle carrying your whole setup. One URL for your team, the full config in one paste, forkable by anyone. See what people are running →