Claude Connectors: Setup and Security

Claude connectors link Claude to external apps and services through the Model Context Protocol, enabling it to search data and perform actions while normally inheriting each user’s existing permissions. Available across web, desktop, mobile, and Claude Code, connectors include prebuilt directory options such as Slack, Google Drive, Gmail, Linear, Asana, and Figma, as well as custom publicly reachable remote MCP servers, with Free users limited to one custom connector. Team and Enterprise administrators can enable connectors organization-wide and control individual tool categories by allowing them, requiring approval, or blocking them, while users generally authenticate separately unless managed authentication is used. The material emphasizes beginning with read-only access, restricting write and deletion capabilities, using OAuth rather than shared API keys, and reviewing each connector’s permissions and tool definitions. It also identifies risks including prompt injection in retrieved content, poisoned or silently changed tool descriptions, excessive permissions, and shared credentials that can bypass per-user access controls; cited research found tool-poisoning attacks succeeded in 22.9% of tests across agent frameworks. Recommended governance measures include limiting custom connector access, monitoring server and tool changes, auditing usage, maintaining an approved-server inventory, and potentially routing custom servers through a centralized gateway for consistent identity, policy, and logging controls.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.