- Sources: Teng Li writeup, HN 49002358
- Summary: A practitioner writeup graded 36 popular Model Context Protocol servers on how usable their tools are for an agent, reporting that about a third scored poorly on issues such as unclear tool descriptions, oversized or unstructured responses that bloat context, and missing error signals. The scoring rubric is the author's own and the sample is self-selected.
- Why it matters: It reframes MCP-server quality as an agent-ergonomics problem, where token-heavy or ambiguous tool outputs directly degrade agent performance.
send feedback on this story