Blog · 16 November 2026 · 7 minute read
What is an MCP server? A plain-English guide for people who build with AI
The Model Context Protocol explained without jargon: what an MCP server is, what tools are, local versus remote servers, and how to stay safe when you connect one.
The short version
An MCP server is a small program that gives an AI assistant new abilities. On its own, a model can only read and write text. Connect an MCP server and it can also do things: look up a record, read a file, check a deployment, start a launch.
MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic in late 2024, that describes how AI tools and outside programs talk to each other. Because it is a standard, a server written once works in every tool that supports it, including Claude Code, Cursor and Codex.
What a server offers
A server describes what it can do in three forms.
- Tools: actions the assistant can take, such as "show the launch plan" or "run the health check". Each has a name, a description and the inputs it needs.
- Resources: information the assistant can read, such as a file or a database record.
- Prompts: ready-made instructions for common tasks.
How a tool call works
When you ask for something, the assistant looks at the tools it has and decides whether one fits. If so, it asks to call it with specific inputs. Your AI tool usually shows you the request and asks for approval, the server runs the action, and the result comes back to the assistant, which uses it in its answer.
The important part is that the server does the work, not the model. The model only decides what to ask for. That is why a well-built server can put firm limits on what is possible, whatever the model requests.
Local and remote servers
A local server runs on your own computer. Your AI tool starts it and talks to it directly. It can reach your files, your terminal and anything else on your machine, and what it handles never has to leave your computer.
A remote server runs on the internet at a web address, and you sign in to it. It suits services that live online anyway. Anything you send to it goes to whoever runs it.
Neither is better in general. For anything that touches your keys or your own files, local keeps you in control. For a shared online service, remote is simpler.
Connecting one
Most tools need one line of configuration: the command that starts the server, or the web address of a remote one. In Claude Code, for example, adding a local server is a single command naming the server and the program to run. After that, its tools appear the next time you start a session.
Staying safe
- Only connect servers from a source you trust. A server can do whatever its tools allow.
- Read what a tool will do before approving it, especially anything that writes, deletes or spends.
- Prefer servers that mark which tools only read and which change things.
- Never give a token to the assistant in the chat. A good server takes credentials through its own secure channel.
- Be careful with content from outside, such as web pages and documents. Text inside them can try to instruct the assistant.
Where Shipyard fits
Shipyard is a local MCP server as well as a command-line program. Connected to your coding tool, it lets the assistant show a launch plan, rehearse it, run the launch and check the result, on your own provider accounts. It has no tool that accepts a provider token: you paste those into Shipyard's own page, and they stay in your computer's keychain. Anything that spends money stops and waits for you.
Launch in an hour, not a weekend
Shipyard connects your accounts once, then sets up the domain, hosting, database, email and payments for every new product.
See how it works