MCP Connectors
SecureAI can connect to Model Context Protocol (MCP) servers so its assistants can call external tools — querying databases, reading tickets, running cloud operations — under governance. Connectors are chosen from a built-in marketplace and configured per user; every tool call still passes through SecureAI’s policy and safety layer. Connectors are managed in Admin → Integrations → MCP (API base/api/connectors).
Transport & authentication patterns
Each connector declares how SecureAI reaches it and how it authenticates. Three patterns are supported:
Some token-based remote connectors (e.g. the Google Cloud family) also carry an OAuth flow so a user can authorize with their own Google OAuth client via a popup, rather than pasting a static token.
Connecting a connector
- Admin → Integrations → MCP.
- Pick a connector from the marketplace.
- Provide its required configuration — an API key/token, or authorize via the OAuth popup for OAuth connectors.
- Save. The connector becomes available to the assistant’s tool router.
Tool safety
Every connector declares blocked tool patterns — destructive operations (e.g.delete, destroy, terminate) that SecureAI refuses to invoke even if the MCP server exposes them. This guardrail is enforced centrally, so a connected tool server cannot be coerced into a destructive action through the assistant.