I graded 36 popular MCP servers on agent usability. A third got a D or F (tengli.dev)

A recent review of 36 popular Model Context Protocol (MCP) servers revealed alarming usability issues, with one-third of the servers receiving a grade of D or F. While compliance with the MCP specification is essential, this analysis highlights that many servers can technically meet the standard yet still fail to effectively serve AI agents. The primary issues stem from a lack of adequate documentation, particularly in parameter descriptions, which leads to incorrect tool selections and inefficient resource use during model operations. The study introduced a new tool, mcpgrade, designed to assess server usability easily, making it possible to obtain quick reports without extensive setup. This finding is significant for the AI/ML community as it underscores that usability is just as critical as technical compliance in the deployment of AI tools. The prevalent epidemic of undocumented parameters complicates agent interactions and can lead to severe malfunctions during real-time operations. The results suggest that focusing on detailed documentation and strict adherence to naming conventions can vastly improve agent reliability and performance, providing a roadmap for developers to enhance their MCP servers. As demonstrated in the review, actively maintained servers often fail in usability compared to archived ones with thorough documentation, emphasizing the need for a paradigm shift where engineering rigor is complemented by clear communication of capabilities and constraints.