PreClick sits between your AI agent and the open web, scanning URLs for phishing, malware, and intent mismatches before navigation happens. It exposes two main tools via MCP: url_scanner_scan for basic threat analysis and url_scanner_scan_with_intent when you want to verify a link actually matches what your agent is trying to do (like confirming a booking URL is really for hotels, not a lookalike). The hosted endpoint at preclick.ai/mcp runs stateless over streamable-http with a 100 request daily trial, no key required. For tighter integration there are standalone Python and Node clients that handle the async polling without touching MCP protocol details. Reach for this when your agent browses autonomously and you need a security gate that understands context, not just pattern matching.
claude mcp add --transport stdio ai.urlcheck-urlcheck-mcp uvx urlcheck-mcp