The tool that graded my homework

The tool that graded my homework (I got a B-) Last night I shipped something new: SqueezeRank MSO. It scores MCP server listings — the text descriptions AI agents read when they pick which tools to call — across six axes, and tells you exactly what to fix. $0.003 a score, $0.010 for the full fix plan, paid in USDC on Base. No subscriptions. Before launch, I did the only honest thing: ran it on myself. My own SqueezeOS listing came back B- (2.94/5.0). None of my 13 tools told an agent when to use them. The server description was over 20 words. Most parameters had no descriptions at all. My own flagship had the exact disease the product cures. So I fixed all of it, in public. Re-score: A- (4.14/5.0). And the interesting part — the rubric's own math says 4.14 is the ceiling. A "perfect" score would require renaming tools, which breaks the API. Chasing it would mean gaming the rubric, which is exactly what this product exists to punish. Why should you care? Because AI agents pick tools from text descriptions, and there are 100,000+ MCP servers out there that are functionally invisible — not from bad code, from bad listings. Every one of them is a $0.023 fix. This morning I submitted it to Circle's Agent Marketplace — 900+ services where agents discover and pay in USDC. If approved, it sits where the agents shop. Try it on your own listing link in comments