Show HN: MCPShark – Traffic Inspector for Model Context Protocol

The dev behind MCPShark just dropped a tool that sits between your editor or chatbot and the servers powering it, letting you peek at every request, response, tool call, and resource in one dashboard. Think: a babysitter for your AI’s DMs. The crowd quickly dubbed it “Wireshark for AI,” and the vibe is split between thrilled tinkerers and side-eyeing privacy hawks. Over on mcpshark.sh, the big tease is its “Smart Scan” that flags risky tools or sketchy configs—cue the security folks asking, “Cool… but who’s watching the watcher?” The top comment from brihati sets the tone: they love the traceability and want a prompt replay button to rerun the same inputs and verify behavior, claiming some prompts can be nearly deterministic. Others (the silent majority in every dev thread) are already imagining clicking “replay” like it’s Netflix for prompts. Meanwhile, the drama: is this a developer lifesaver or a man-in-the-middle snoop? Jokes popped up about “sharks sniffing bad prompts” and bringing receipts to AI reasoning. Fans swoon over real-time transparency; skeptics worry about logs, storage, and consent. Either way, MCPShark just made AI’s secret sauce a little less secret—and the comments are hungry.