Is the MCP protocol open source? Do my own MCP servers have to be?

The MCP specification and the official SDKs (TypeScript, Python, Kotlin, Swift, C#) are open source under the MIT license at github.com/modelcontextprotocol. Individual MCP servers are not required to be open source - that is entirely the maker's call. You can list and launch a closed-source or commercial MCP server on LaunchMCP exactly like an open source one.

Anthropic introduced MCP in November 2024 as an open standard. It has since been adopted by every major AI assistant including ChatGPT, Cursor, Windsurf, and Cline.

Do I need to know how to code to use an MCP server?

No. Installing an MCP server is usually a one-line config change in your AI client. Building a new MCP server requires coding, but installing and using one does not.

What is the difference between an MCP tool, resource, and prompt?

Tools are functions the AI can call (e.g. query_database). Resources are read-only data the AI can attach as context (e.g. file contents). Prompts are reusable templates the user can invoke. Most servers expose tools; resources and prompts are optional.

Guide

What is MCP? A complete guide to the Model Context Protocol

MCP, short for Model Context Protocol, is an open standard that lets AI assistants like Claude, ChatGPT, Cursor, and Windsurf connect to external tools, data sources, and APIs. It was introduced by Anthropic in November 2024 and has since become the default way to give an AI agent real-world capabilities.

This guide explains how MCP works, why it matters, the clients that support it, and how to install or build your own MCP server. If you just want to find good MCP servers, head to the directory. If you are about to ship one, read how to launch your MCP server.

The 30-second definition

MCP is to AI agents what USB-C is to laptops: a single, standard plug that replaces a dozen custom adapters. Before MCP, every team that wanted Claude to talk to Postgres or GitHub had to hand-roll a tool integration. With MCP, anyone can write a server once and every MCP-compatible client can use it.

How MCP works

MCP is a client-server protocol over JSON-RPC. There are three roles:

Host: the AI assistant you talk to (Claude Desktop, Cursor, ChatGPT, Windsurf).
Client: a thin layer inside the host that manages one connection to one MCP server.
Server:the program that exposes tools, resources, and prompts. This is what people mean by “an MCP server.”

On startup, the client asks the server “what can you do?” and the server replies with a list of tools, each with a name, description, and JSON schema for its arguments. When the assistant decides to call a tool, the client forwards the call, the server runs it, and the result comes back as structured data the assistant can reason over.

Two transports exist today. STDIO runs the server as a local subprocess of the host (fast, private, no network). HTTP (with Server-Sent Events) runs the server as a remote service the host connects to over HTTPS (good for shared and team servers).

Why MCP matters

Three reasons MCP became the standard so quickly:

Write once, run in every assistant. A Postgres MCP server works in Claude, Cursor, Windsurf, and ChatGPT without a single line of client-specific code.
The user controls the surface area. The assistant cannot call a tool the user did not install. Permissions live with the user, not the model provider.
Structured tool results.Unlike text-only “plugins,” MCP results come back as typed JSON, so the assistant reasons over them instead of parsing prose.

Which clients support MCP?

As of 2026, MCP is supported by:

- Claude Desktop and Claude Code (Anthropic)
- Cursor
- Windsurf (Codeium)
- ChatGPT (via the Apps SDK)
- Cline (VS Code extension)
- Continue, Zed, and a growing list of IDE plugins

Each server on LaunchMCP is tagged with the clients its maker has verified, so you can filter the directory by your assistant.

What can an MCP server do?

Anything you can expose as a tool. The popular categories on LaunchMCP today:

Databases: Postgres, MongoDB, Redis, BigQuery, Snowflake.
Developer tools: GitHub, Linear, Jira, Sentry, Vercel.
Design and content: Figma, Notion, Google Docs, Airtable.
Browser and search: Playwright, Perplexity, web search.
Filesystem and shell: read files, run scripts, manage Git.

How to install an MCP server

Every server page on LaunchMCP includes a one-line install command for each supported client. The general flow:

Pick a server from the directory and open its detail page.
Copy the install command for your client.
Paste it into your client's MCP config (or run the CLI command).
Restart the client. The server's tools appear in your next conversation.

How to build your own MCP server

The official SDKs (TypeScript, Python, Kotlin, Swift, C#) handle the protocol for you. The minimum viable server is about 40 lines:

Install the SDK (npm i @modelcontextprotocol/sdk or pip install mcp).
Create a server, register a tool with a name, description, and Zod or Pydantic schema for its arguments.
Implement the tool handler to do the actual work.
Wire up the STDIO transport and run the process.
Test it locally by adding it to Claude Desktop's config.
When ready, launch it on LaunchMCP to reach the developers actively looking for new servers.

Ready to find one or ship one? Head to the MCP server directory or read how to launch. Still have questions? The full FAQ covers everything else.