Still relying on HTTP 200 OK to monitor your AI agent's MCP server? Openstatus just dropped a tool that exposes why your basic ping is completely useless.
So you're building an MCP (Model Context Protocol) server to plug into Claude or Cursor, huh? You hook it up, hit your health endpoint, see that sweet HTTP 200 OK, and call it a day. Hate to break it to you, but you’re living on pure copium. When your AI agent trips over a broken JSON-RPC handshake, that 200 status means absolutely squat.
The folks over at Openstatus (Tibo and Max) apparently got fed up with debugging this exact flavor of nonsense, so they dropped the "MCP Server Health Check" on Product Hunt.
TL;DR: Standard HTTP pings are garbage for AI endpoints. If your JSON-RPC handshake fails or if the tools/list comes back completely empty, your agent breaks, end of story.
This open-source, zero-install tool was built to test your endpoints exactly how a real AI client would. It ignores your useless pings and instead:
initialize, ping, and tools/list sequence.The dev community chimed in, and the vibe is a mix of "finally" and "yeah, but...":
trekh went straight for the jugular: "When you say 'like a real ai client', does it actually call a tool, or stop at the handshake? Bugs hide in the actual execution." The maker, Tibo, had to admit they're still refining the final tool execution step.alpertayfurr and ozandag united in laughter at how utterly useless a 200 status is for MCP. It’s those edge cases in the actual tool flow that will nuke your server.ansari_adin pointed out the "Wild West" state of MCP right now, where servers are hastily built on ancient draft specs. They suggested adding a compliance badge so developers know a server isn't just faking its uptime.Look, if you’re building AI tools or AI agents, it's time to throw your old REST API CRUD mindset out the window. Protocol-level validation is the new baseline. Stop relying on basic uptime monitors before your production agent starts hallucinating or crashing just because it couldn't fetch its tool list.
Test it like it's actually going to be used, not just pinged by a dumb monitor. Don't push to prod on a Friday without protocol-level testing, unless you enjoy 3 AM hotfixes.