Executive brief
The Model Context Protocol (MCP) Registry is a directory for AI-related server components. A vulnerability in the registry's web interface allows malicious publishers to inject malicious scripts into the catalogue. If a user views a poisoned entry, the attacker could steal authentication tokens, redirect the user to malicious sites, or perform actions on the user's behalf within the registry.
Technical details
A stored cross-site scripting (XSS) vulnerability exists in the MCP Registry catalogue UI due to improper output encoding of the `websiteUrl` field. While the application uses a homegrown `escapeHtml` helper, this helper only sanitizes characters for text content (like `<` and `&`) and fails to escape double quotes (`"`). An attacker with publishing privileges can provide a URL containing a double quote to break out of the `href` attribute in the `ui_index.html` template. This allows the injection of arbitrary HTML attributes, such as `onmouseover`, which execute JavaScript when a user interacts with the malicious entry. The existing Content Security Policy (CSP) fails to mitigate this because it includes `'unsafe-inline'`. The issue is fixed in version 1.7.7 by improving validation and encoding.
Affected products
- modelcontextprotocol registry < 1.7.7
Timeline
- 2026-05-04: disclosed: Vulnerability reported to vendor
- 2026-05-08: advisory: GitHub Advisory published
- 2026-05-08: patched: Fix released in version 1.7.7