Mcp Server

Think of this as a bouncer for your LLM's front door. It filters incoming prompts for three specific threats: prompt injection attempts, evasive spacing tricks (like i n s e r t i n g spaces to bypass filters), and obscene content. The screening happens before anything hits your model, which means you catch bad input early rather than trying to sanitize responses after the fact. The repository doesn't expose detailed APIs in the visible source, but the core value is clear: you drop this server into your MCP setup and it acts as middleware between user input and your expensive LLM calls. Useful if you're building customer-facing agents or chat interfaces where you need a first line of defense against abuse and jailbreak attempts.