InboxWall: Local MCP Server Sanitizes Email HTML to Block Prompt Injection
InboxWall is an open-source Model Context Protocol server that cleans email HTML before an AI agent reads it, guarding against prompt injection. Attackers can hide instructions in emails using invisible CSS text, manipulative Unicode characters, or HTML comments that only appear in raw source. The tool exposes a single function that takes raw email HTML and returns plain text with styles, scripts, hidden characters, comments, and risky attributes removed. It runs entirely on the user's own machine, so email content is never sent to a third-party service, though the agent must manage the server as a separate process. The project's core design choice is to treat sanitization as a security boundary outside the language model's trust zone.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in