Transparency log

Security Findings

Automated scans occasionally flag patterns on this site. This page lists the findings that have been reviewed, the current status of each, and the reasoning behind the decision. It is maintained by the site owner and is not a certification.

Public MCP server exposed without authentication

Lovable security scanner · Severity: Informational · Reviewed July 2026

Ignored — accepted risk

Finding

The Model Context Protocol server mounted at /mcp is reachable without an authenticated session, allowing any external client to enumerate its tools.

Justification

This is the intended design. The MCP server exposes read-only editorial content (dispatches, chapters, field notes) that is already published on the public site. It handles no user data, performs no writes, and holds no privileged credentials. Public reachability is what allows third-party AI clients to browse the journal, which is the whole point of publishing it.

id: app_mcp_public_unauthenticated

subscribers table has no SELECT policy

Lovable security scanner · Severity: Low · Reviewed July 2026

Ignored — accepted risk

Finding

The subscribers table has RLS enabled but no SELECT policy for anon or authenticated roles, so it cannot be read from the client.

Justification

This is fail-closed by design. The subscribers table only accepts newsletter INSERTs from the client; nobody is meant to read from it via the Data API. Administrative reads happen through the service-role key on the server. Adding a SELECT policy would widen the surface area rather than reduce it.

id: subscribers_no_select_policy