Top AI Product

Openstatus shipped an MCP Health Checker — a small but useful piece of plumbing for the now-sprawling Model Context Protocol ecosystem. It tests an MCP server the way a real AI client would, not by pinging a port and calling it healthy.

## A real handshake, not a TCP check

The tool speaks JSON-RPC and runs the same spec-defined sequence an MCP client follows on first connection: initialize, ping, tools/list. A passing check confirms the server is actually reachable, speaks the handshake correctly, and exposes the tools it claims to — which is what AI clients like Claude Desktop, Claude Code, Cursor, and Continue need to see before they’ll work with it. A common failure it surfaces: a server returning 401 without a WWW-Authenticate header is non-compliant, and most clients silently fail to authenticate against it.

## Why it matters

MCP went from one protocol to “every major AI client supports it” in months. That growth happened faster than tooling for verifying servers actually behave correctly. A server that mostly works is one of the harder failure modes to debug — half your clients can call it, half can’t, and the logs say nothing useful. Treating MCP servers like HTTP endpoints with proper health checks is the obvious move, and it’s surprising it took this long for someone to ship one as a focused tool.