Hey Product Hunt! I'm Gautam. We build at Eulogik, and this is OpenTrustBench. Short version: it's a free scanner that grades AI agents and MCP servers A through F. npx @opentrustbench/cli scan . Ten seconds later you have a Trust Card with a letter grade, OWASP-mapped findings, and SARIF output for CI. Why we built it: everyone's installing MCP servers that get shell access, file access, API keys… without reading them. We were doing it too. Felt like the browser-toolbar era, so we fixed my own habit first. The part I'm proudest of: it's honest by design. Static analysis only, nothing executes, no invented findings. Attack mode is heuristic, eval is simulated, and the Trust Card is evidence, not a certification. We print all of this in the docs instead of hiding it. And my favorite detail: our own repo grades F, because it ships a deliberately vulnerable test fixture (a scanner that passed its own ammunition would be worthless). The CLI we actually ship grades A. We also scanned 53 public MCP servers, average C, 36% D or F, all published with full evidence. Free forever, Apache-2.0, zero telemetry. If we ever charge for anything, it'll be human expert reviews, the scanner itself stays free. Try it on something you installed recently and I'd genuinely love to hear: did the grade surprise you, and what rules are we missing? I'll be here all day. Website: opentrustbench.com · Registry: opentrustbench.com/r/ · GitHub: github.com/eulogik/OpenTrustBench

OpenTrustBench

A trust score for every AI agent

A trust score for every AI agent