a huge reason we use @newitemco is because of their 10/10 MCP (and also their awesome team - s/o @akguthal). it's the same story for @meetgranola and @linear.
a year ago, fair. MCP was iffy. auth was duct tape and barely any clients supported it. since then the spec grew up and every major AI client adopted it.
now it's part of my daily (let's be honest, minute-to-minute) workflow.
now's the time to build that mcp you've had in your backlog for months!!
Failing marketplace review is so common it's becoming a rite of passage, but it doesn't have to be. Here's what your MCP app has to survive to make it into the directories:
1/ Initialization behavior: your server has to handshake cleanly every single time. Flaky init is an instant reject.
2/ Tool schemas: every tool needs a schema an LLM can route on. Vague descriptions and loose types get flagged automatically.
3/ Error handling: structured errors only. Silent failures and raw stack traces leaking to the client both fail review.
4/ Metadata quality: name, description, icons. reviewers judge it like a listing page, because that's what it becomes.
5/ The biggest piece: ChatGPT and Claude run your tools differently, and apps that render in one break in the other.
We actually turned this whole audit into one click. Every check, pass/warn/fail, plus the golden fix prompt you can paste straight into your agent for anything that breaks: https://t.co/pyvu0V19o3.