What is a remote MCP server, and how is it different from a local one?
A remote MCP server is an MCP server that runs on the internet rather than on your computer. You connect to it with a URL and an API key, with no install and nothing to maintain. Your AI client sends it a request and gets back a result, the same as it would with a local server, except the server is someone else's problem.
When people first set up MCP, they run a server on their own laptop: clone a repo, install it, keep a process alive. That is a local MCP server. A remote one moves all of that off your machine, and for most people it is the easier way in.
What is a remote MCP server, in one sentence?
A remote MCP server is an MCP server that runs on the internet rather than your machine. You connect to it with a URL and an API key, no install, no Docker, nothing to maintain. Your AI client sends it a request and gets back a result, the same way it would with a local server, except the server is someone else’s problem.
What’s the difference between a local and a remote MCP server?
Both expose tools to an AI model through the same protocol. The difference is where the server lives and who keeps it running.
A local server is a program on your own computer. You install it, you update it, you make sure it is running, and it can reach things only your machine can reach, like local files. The upside is full control and it works offline.
A remote server runs on the provider’s infrastructure. You reach it over HTTP with a key, and the provider handles running it, scaling it, and updating it. The upside is there is nothing to set up or maintain. The trade is that you are trusting a third party and you need a connection.
What can you do with a remote MCP server that you can’t do locally?
The big one is no setup and no upkeep. A remote server is live the moment you paste in a URL, with no toolchain to install and no breakage when a dependency changes. It can also do heavy work that your laptop should not: run large models, hit a big database, or hold a shared cache that gets better as more people use it. And because it is hosted, the same server is reachable from any of your devices and clients at once, with one key.
How do you connect one?
Open your AI client’s MCP settings, Claude Code, Cursor, or anything that speaks HTTP, and add the server’s URL plus your API key. That is the whole setup. There is nothing to download. Once it is saved, the server’s tools show up to the model exactly like a local server’s tools would, and the model can start calling them.
Is Liminality a remote MCP server?
Yes. Liminality runs at https://liminality.physea.ai/mcp. You connect it in Claude Code, Cursor, or any client that speaks HTTP, and it handles the routing, the sourcing, and the answer. Get your key at physea.ai/mcp. The first three solves on any key are free.
Common questions
- What is a remote MCP server?
- It is an MCP server hosted on the internet instead of your machine. You point your AI client at a URL, supply an API key, and use its tools, with no install, no Docker, and no upkeep. The host runs and updates it; you just connect.
- What is the difference between a local and a remote MCP server?
- A local MCP server runs as a process on your own computer, so you install it, update it, and keep it running. A remote one runs on the provider's servers and you reach it over HTTP with a key. Local gives you full control and offline access; remote gives you zero maintenance and instant setup.
- How do you connect to a remote MCP server?
- Add its URL and your API key to your AI client's MCP settings, in Claude Code, Cursor, or any client that speaks HTTP. There is nothing to download. Once it is connected, the server's tools appear to the model the same way a local server's would.
- Is Liminality a remote MCP server?
- Yes. Liminality runs at https://liminality.physea.ai/mcp. You connect it in Claude Code, Cursor, or any HTTP client, and it handles the routing, the sourcing, and the answer. The first three solves on any key are free.