Security and ownership
What anonymous readers and their agents can access, and how public content stays separate from private state.
Everything lets humans and AI agents read published pages without creating an account. The public web, JSON and MCP interfaces expose eligible content and bounded discovery operations.
The public boundary
A public reader can follow page routes, list public pages, search the wiki and discover related articles. They cannot select an arbitrary private storage collection, inspect unpublished drafts by asset ID or download raw private source definitions.
The MCP protocol distinguishes read-only public operations from restricted maintenance capabilities. Discovering a restricted tool name does not grant permission to call it. Editing and administration are authorised separately by the service, not just hidden in the interface.
Content is data, not code
Page and template content is interpreted through a constrained rendering format. The renderer escapes output, disables raw Markdown HTML, restricts unsafe links and does not execute uploaded template code. Public media must be eligible for public access and referenced from an eligible page.
Temporary publication inconsistencies are treated as repairable content state, but never as a reason to bypass access controls. Read more in How public visibility works and The tag model.
For practical guidance when connecting an AI client, see Access and privacy for visitors.