MCP FAQ: 12 answers about the Model Context Protocol
Everything you need to know about the Model Context Protocol: how MCP servers work, how to install them, which clients support them, and how to build your own.
Basics
4 questions about basics.
What is MCP?
MCP (Model Context Protocol) is an open standard that lets AI agents call tools, read resources, and use prompts exposed by external servers. It is a JSON-RPC based protocol that runs over stdio, HTTP, or Server-Sent Events.
What is an MCP server?
An MCP server is a small program that implements the Model Context Protocol. Most MCP servers wrap a CLI, a database, a SaaS API, a local file system, or a remote service so the agent can talk to it through a single, standardized interface.
Is MCP free to use?
Yes. The MCP specification, SDKs, and reference servers are open source and free to use. Some MCP servers wrap third-party services that may have their own pricing (for example, GitHub, Stripe, or Notion), but the MCP layer itself does not charge a fee.
Who maintains MCP?
MCP was introduced by Anthropic in late 2024 and is now an open standard with reference SDKs available in Python, TypeScript, Go, Rust, and Java. The community maintains a long list of official and community servers.
Installation
3 questions about installation.
How do I install an MCP server?
Find a server in the MCP Haus directory, copy the install command on its detail page, and register the command in your AI client's MCP configuration. For Claude Desktop, edit claude_desktop_config.json. For Cursor, use the settings UI. Most setups take under a minute.
Do I need Node.js to run MCP servers?
Many MCP servers are written in TypeScript and run via npx, which ships with Node.js. If you prefer Python, uvx and pipx are common alternatives. Docker is a good choice if you want to isolate the server from your host system.
Can I run multiple MCP servers at once?
Yes. MCP clients support multiple simultaneous servers. A typical setup includes a GitHub server, a Postgres server, a Filesystem server, and one or two SaaS-specific servers like Notion or Linear. The agent uses whichever tool fits the request.
Compatibility
3 questions about compatibility.
Which AI clients support MCP?
Claude Desktop, Cursor, Continue, Windsurf, Cline, OpenHands, and Claude Code all support MCP. The list is growing because MCP is a standard rather than a single-vendor API.
Can I use MCP with OpenAI or other model providers?
MCP is model-agnostic. The client implements the protocol and the model can be any LLM the client supports. Some clients are Claude-first (Claude Desktop, Claude Code) while others (Cursor, Continue) work with multiple model providers.
Do MCP servers work on Windows, macOS, and Linux?
Yes. MCP is cross-platform. Some servers use shell-specific commands (like wget or curl), but the protocol itself works on every major operating system.
Building MCP servers
2 questions about building mcp servers.
How do I build my own MCP server?
Pick a language (Python and TypeScript are most common), use the official SDK, expose a list of tools, and start the server over stdio or HTTP. The MCP docs include a quickstart that walks you through a "hello world" server.
Can I publish my MCP server on MCP Haus?
MCP Haus currently indexes the most popular public MCP servers. If you have built a server that other developers would find useful, you can open an issue on the project repository and we will consider adding it to the directory.