An MCP connector is an integration that lets an AI assistant reach a remote tool or service, such as email, calendars, ticketing systems, files, or internal APIs. It sits between the model and the target system, translating requests into actions the remote service can understand.
In cyber security, connectors matter because they extend the assistant’s trust boundary. If a connector has broad permissions, a prompt injection, stolen token, or misconfigured policy could let an attacker read data, move laterally, or trigger unwanted actions through the AI workflow. Defenders should treat connectors like privileged integrations: apply least privilege, scope access tightly, log activity, and revoke tokens quickly when users change roles or leave. When governed through identity providers and enterprise policy, MCP connectors can be easier to audit and control, but only if authorization and token handling are kept strict.