What is a remote MCP server?
A remote MCP server is a Model Context Protocol server your AI agent reaches over the internet at a URL, instead of starting it as a program on your own machine. It speaks the Streamable HTTP transport, and most ask you to sign in with OAuth the first time, so nobody has to copy an API key.
Part of Glossary: the terms behind MCP, coding agents and shared pages, defined plainly.
Your agent's MCP client connects to one HTTPS address, such as https://app.intraface.cc/mcp. Nothing runs on your machine.
The first time, the server asks you to sign in. Your browser opens, you click Allow, and the client keeps the token.
From then on every request goes over HTTPS and the server answers, streaming its reply when it has a lot to say.
Every client that knows the address gets the same tools: Claude Code, Codex, Cursor and the rest, on every teammate's laptop.
Source: MCP specification: transports
Remote versus local MCP servers
A local MCP server is a program your client starts on your own machine and talks to through its standard input and output. It can read your files, open your browser, or run commands, because it lives where you do. A remote server is already running somewhere else. Your client sends it messages over HTTPS, the same way your browser talks to a website, and the server does its work on its own infrastructure.
That difference decides most things about it. A remote server is installed once, by whoever runs it, and updated in one place. Its sign-in and permissions belong to the service, not to a key in a config file on your laptop. And it works the same from every client, so a team can share one address instead of each person setting up a copy.
When to use a remote server
Use a remote server when the data lives in a service many people share: a board, a ticket tracker, a document store, a design tool. Use a local server when the tool needs your machine, like a filesystem server or a browser you control. Many setups use both: a local server for your files and a remote one for the things your team works on together.
Common mistakes
- Treating the URL as the whole setup. Most remote servers answer the first request with a sign-in step. If your client shows "needs authentication", that's expected; finish the sign-in and try again.
- Pasting a key where the server wanted OAuth. If the server supports browser sign-in, use it. A token in a header is for agents that can't open a browser.
- Expecting a remote server to see your files. It can't. Anything it works on has to be sent to it, which is why agents pull a board first and publish back afterwards.
FAQ
- Is a remote MCP server the same as an MCP connector?
- Mostly. Claude.ai and other apps call a remote MCP server you add to them a connector. Underneath it's the same thing: a URL that speaks MCP over Streamable HTTP.
- Do I need to run anything on my machine to use a remote MCP server?
- No. You give your client the URL and sign in. A local server is the opposite: your client starts it as a program, so it has to be installed where the client runs.
- Is SSE still supported for remote MCP servers?
- HTTP+SSE is deprecated. Some clients still fall back to it for older servers, but new servers offer Streamable HTTP, which can still stream responses as server-sent events when it needs to.
Stop sending HTMLover Slack.
Create an account in a minute, connect your coding agent, and make your first board.