If you've spent any time with AI assistants in the last year, you've probably noticed that the gap between "AI that answers questions" and "AI that actually does things" keeps closing fast. A big part of why is MCP.
The USB-C Analogy
MCP stands for Model Context Protocol. It's an open standard — originally developed by Anthropic and now adopted across the industry — that gives AI applications a standardised way to connect to external systems.
Think of it like USB-C for AI. Before USB-C, every device had its own connector. You needed a different cable for your laptop, your phone, your monitor. USB-C unified all of that into one standard.
Before MCP, every AI app had to build its own integrations: custom code to talk to Google Drive, custom code to query a database, custom code to trigger a GitHub action. Each integration was a one-off. MCP unifies all of that.
With MCP, you build the integration once — as an MCP Server — and any MCP-compatible AI client can use it.
The Three Primitives
MCP has three core building blocks:
MCP Servers expose tools, resources, and prompts. A tool might be "search the web" or "query my database". A resource is a document or data source the AI can read. A prompt is a reusable instruction template. Thousands of MCP servers already exist: for Slack, GitHub, Notion, PostgreSQL, web search, local files, and hundreds more.
MCP Apps are the newer, richer cousin. Instead of returning plain text, an MCP App can return an interactive HTML interface — a dashboard, a form, a 3D visualisation — that renders directly inside the AI client. Claude Desktop, VS Code Copilot, and Goose already support them. You get a full interactive UI without leaving the conversation.
Agent Skills are portable packages of procedural knowledge — think SKILL.md files that tell an AI agent how to do a specific task reliably. Skills are supported by Claude Code, Cursor, Gemini CLI, GitHub Copilot, and 30+ other agent tools. Write a skill once, deploy it everywhere.
Why Developers Care
The core value proposition for developers is simple: build once, work everywhere.
An MCP server you write today will work with Claude, ChatGPT, VS Code Copilot, Cursor, Goose, and any other client that implements the protocol. You're not locked into one AI vendor's API.
For teams, this means your internal tools — your data warehouse, your incident management system, your design files — become AI-accessible without rebuilding for every assistant your engineers want to use.
For indie developers, it means you can publish an MCP server or app to a directory and immediately reach users across every major AI platform.
The Ecosystem Is Moving Fast
MCP was open-sourced in late 2024. By early 2026, it's become the de facto standard for AI integrations. The spec is evolving — MCP Apps shipped as an official extension, Agent Skills have been adopted by every major coding agent — and the pace isn't slowing.
The challenge now isn't "does MCP exist?" It's "how do I find the right MCP server for my use case?" and "how do I know which one is well-maintained?"
That's exactly what this directory is for.
Browse the Directory
We track MCP Apps, MCP Servers, and Agent Skills in one place — with descriptions, install instructions, and compatibility information so you can find what you need without trawling GitHub.