July 22, 2026
Bots got grades. Comments chose violence
I graded 36 popular MCP servers on agent usability. A third got a D or F
Even the scorecard link broke, and commenters pounced on the mess
TLDR: A developer found that many popular tools built for AI assistants are surprisingly hard for the AI to use, often because the instructions are missing or unclear. Commenters split between laughing at the broken results link, dunking on the post as “slop,” and arguing it exposes a real design problem.
A developer tested 36 popular tools that help AI assistants connect to apps and services, and the report card was brutal: about a third flunked on basic usability, including big-name official offerings from MongoDB, Notion, and Airtable. The big reveal was almost embarrassingly simple: many of these tools technically "work," but they don’t explain themselves well enough for an AI to use them properly. In plain English, the bot sees buttons with vague labels and no instructions, then presses the wrong one. Ouch.
But the comments quickly turned this from a nerdy audit into a full-on popcorn thread. One of the first reactions was pure comedy: the link to the results table was broken. For critics, that was instant ammo. Others went straight for the jugular, dismissing the whole thing as "AI slop" and "just slop," which is basically the internet’s version of throwing tomatoes. Meanwhile, a more thoughtful camp argued the real issue isn’t just bad descriptions, but bad design: huge flat lists of actions are confusing, and smarter grouping matters more than raw size.
There was also a bigger philosophical hot take bubbling underneath: have we all been pretending data can explain itself? One commenter argued this messy moment proves machines need richer descriptions and better semantics, not just piles of code and spreadsheets. So yes, the article says many AI-friendly tools are failing. But the real drama is that the community can’t decide whether this is a useful wake-up call, a design manifesto, or just another overhyped AI report with a broken link.
Key Points
- •The article says MCP specification compliance does not ensure that AI agents can effectively select and use tools exposed by an MCP server.
- •The author created mcpgrade, a command-line scorecard tool, and used it to analyze 36 popular MCP servers in a static snapshot.
- •Fifteen of the 36 servers received an A grade, while 11 received D or F; Stripe and Supabase were excluded because they could not be scanned with dummy credentials.
- •The article identifies missing parameter descriptions as the main issue in low-scoring servers, especially under rule D004.
- •The author argues that usability depends heavily on documentation quality, naming, and schema design, and that these are separate from protocol compliance.