Before you start
This guide assumes you’ve read MCP, Explained Simply. In short: an MCP server gives an AI assistant real tools to call. Here we connect one.
Note: AI assistant settings change often. Treat the menu names below as a map, not a transcript — the exact labels may differ in your version.
What you need
- An AI assistant that supports MCP (recent versions of Claude and ChatGPT do).
- An MCP server to connect to — either one you run locally or a hosted one.
- The server’s connection details (a command to run it, or a URL).
The general shape of it
Every assistant follows the same three steps, even when the menus differ:
- Open the assistant’s tools / connectors settings. Look for a section named something like Connectors, MCP servers, Tools, or Integrations.
- Add the server. Provide either the local launch command or the hosted URL. Give it a clear name you’ll recognize.
- Approve the tools. The assistant will show you which tools the server exposes and ask permission. This is the safety checkpoint — you stay in control of what it can call.
How to tell it’s working
Ask the assistant something that only the tool can answer — current, specific, not in its training data. With our Pet Recall Watcher, for example:
“Use the recall tool to check whether [product] has any recent recalls.”
If it’s connected, the assistant will call the tool and answer from live data, often telling you it’s checking the tool first. If it answers vaguely or says it can’t access current data, the connection isn’t active yet.
Common gotchas
- The server isn’t running. Local servers must be launched before the assistant can reach them.
- Tools weren’t approved. Re-open the connector settings and confirm the tools are enabled.
- Wrong URL or command. Double-check the connection details against the server’s documentation.
Where to go next
- GitHub for Non-Developers — if running a server from a repo feels intimidating, start here.
- Discuss a project — want a custom MCP server for your own tools?
Want to put this into practice?
See the tools we've built, or tell us what you're trying to solve.