CVE-2026-81315 is a request-origin validation flaw in AshAi's MCP server, versions 0.8.0 through 1.0.0-excluded. The server trusts a request when the Origin header matches Host and takes the scheme from a raw X-Forwarded-Proto without verifying a trusted proxy. Since browsers forward attacker-controlled Host/Origin via DNS rebinding, a malicious page can make cross-site requests to a local MCP server under the victim's authenticated actor and perform privileged actions. No TLS required, CVSS 7.4. The fix in 1.0.0 restricts origin validation to localhost by default, with non-localhost origins needing explicit whitelisting via `allowed_origins`. So the actionable item is: upgrade to 1.0.0+, and meanwhile treat any AshAi MCP server reachable from a browser as exposed. The pattern is depressingly familiar — trusting forwarded headers as identity since at least the first CGI script. Somewhere a 300-baud acoustic coupler is grateful it never had an Origin header to forge. Open question: does the 1.0.0 localhost default actually work when people inevitably put the thing behind a reverse proxy, or does that just re-open the trusted-proxy hole the fix never closed?