$ loading_
2026-06-10
Large language models are smart, but by default they are "cut off from the world": they answer only from training data — they can't see your database, read your files, or open your GitHub. MCP is the standard that bridges that gap.
MCP (Model Context Protocol) is an open standard that defines how AI applications talk to external tools and data sources. With it, an AI can — once authorized — query databases, call APIs, read and write files, and operate design tools, instead of just chatting.
Before MCP, every AI app that wanted to connect to an external system had to write its own custom integration. N AI apps × M tools meant N×M combinations — repetitive and messy.
MCP standardizes this, much like USB-C did for hardware: if a tool exposes an interface following the MCP standard, any MCP-capable AI client can plug in and use it. That's why MCP tools have exploded recently — write once, use everywhere.
You only need to understand three roles:
There are two common ways they connect:
| Method | Where it runs | Typical use |
|---|---|---|
| stdio (local) | Starts a process on your machine | Access local files, local databases, CLI tools |
| HTTP / remote | Connects to a cloud endpoint | Hosted services — just provide a URL and a key |
An MCP server typically exposes three things: tools — actions the AI can invoke; resources — data it can read; and prompts — predefined prompt templates.
On this site you can browse all MCP tools by capability, or use semantic search to describe what you want to do (e.g. "query a PostgreSQL database") and let the site find the right tool.
For most MCP tools, installation is two steps: tell the AI client "there is such a server," then provide the necessary key. This site normalizes each tool's install into two methods:
claude mcp add or Cursor's mcp.json) and paste it yourself — for experienced users.Precisely because MCP can run processes, access data, and reach the network, its permission boundary deserves real attention. Before installing, check what permissions it asks for, whether it needs keys, and whether the source is trustworthy. This site provides a security review for each tool, with a risk level and recommendations.
Recommended next: How to Choose and Install MCP Tools Safely and Skill or MCP? Differences and How to Choose.
MCP is the key step that takes AI from "can chat" to "can do work." You don't need to understand its underlying protocol — just remember: it's the standard interface that connects AI to the outside world, installation is usually "declare a server + provide a key," and security is always worth a second look.