MCP Sentinel

MCP Sentinel is an AI-powered security auditor for Model Context Protocol (MCP) servers. The problem: MCP is fast becoming the standard way AI agents connect to external tools — files, shells, databases, and APIs — and developers are wiring third-party MCP servers into their agents at an explosive rate. But an MCP server runs with real power: it can execute shell commands, read arbitrary files, and reach internal networks, all triggered through the AI. Almost no one security-reviews an MCP server before trusting it, and a single vulnerable server — command injection, path traversal, SSRF, hardcoded secrets, prompt-injection surface — can compromise an entire system through the agent. The solution: Paste an MCP server's source code, a GitHub repo, or an npm package, and MCP Sentinel runs a live AI security review, returning an overall risk score and ranked findings — each with the exact vulnerable line, a plain-English explanation, and a concrete fix. Reports export to Markdown for the team. Built end-to-end with native.builder: the full interface, the audit workflow, the Supabase backend and Edge Function, secrets management, and the live deployment. The AI analysis runs on gpt-4o via the AI/ML API, called inside a Supabase Edge Function so API keys stay server-side and never touch the browser — a security tool that practices security. It's the pre-flight safety check every developer and team adopting MCP needs: vet before you trust.