Can Your Agent Execute Anything? Authorization for Backstage MCP Actions - Masato Kozuka

Why watch

This session addresses the authorization gap in Backstage's MCP integration, where authenticated AI agents and IDEs can currently execute any exposed action without proper access control. The speaker proposes a refined permission model separating execution from visibility, implementing fail-closed policies for destructive actions, and integrating with Backstone's existing permission framework. This is essential for platform engineers and DevOps teams deploying Backstage in production environments with AI agents.