A connector lets an agent reach Gmail, Slack, Google Drive, QuickBooks, or another system instead of staying trapped in chat. It shows up as a tool the agent can use, sometimes built through MCP, sometimes riding an API underneath. “Read my emails and flag the ones needing a response” is a connector-style task. The access level is the real story, so always check it.
A keycard is the everyday version. You give an employee a keycard for one office suite. It opens the rooms they’re allowed to enter and nothing else. Some keycards open one room, some a whole floor, some only during business hours. A connector is that kind of access for an AI tool.
Why you care
Without connectors, an agent works only with what you paste in. With them, it works closer to the real source of truth. Access is never neutral, though: the moment AI touches your real tools, it gets more useful and more sensitive at once. Ask every time: what app does this open, what can it read, what can it write, and does it need approval before it acts?