A connector is the access you give an AI chat to one of your apps or services, like Gmail, Google Drive, Google Calendar or Slack, so it can read your data and do things for you. It works on MCP. You authorize it with your account and decide what it is allowed to do.
A question from the app. Pick an answer.
You connected the MCP server and it works. What does that tell you about safety?
MCP has authorization mechanisms, but connecting certifies nothing. Check the server and what access it has.
Connecting, signing in and authorizing are different steps. Working only means the connection is fine.
A famous server with full access can still delete everything. What matters is what you allow.
A connector saves you from copying the data from your other apps into the chat. You can ask "summarize this week's supplier emails" or "find the Miga proposal in Drive". Claude, ChatGPT and others offer something equivalent, though each names and limits it in its own way.
A connector is what you see: a button to connect your account. Underneath sits an MCP server, the program that offers those actions. A plugin is something else: it can bring connectors, but also instructions and other pieces.
What the chat sees and can do comes from your account, not from what you ask. If you connect your email, someone else's message with a hidden order may try to trick it: that is prompt injection. So give only what is needed and let it ask for your approval before sending or deleting.
Before accepting, check what it asks to be able to do, and disconnect what you no longer use. And only connect services you trust, especially a connector someone built on their own.
"I connected Drive and now it reads everything." That depends on what you let it do, and the access is limited by the connected account or credential. "If it works, it is safe." Connecting does not certify the service: check who maintains it and which permissions it asks for. And if you no longer use it, disconnect it.