Vague MCP Tool Descriptions Cause Agents to Pick the Wrong Tool

An MCP server can expose every operation correctly yet still fail in practice, because models choose tools based on the names and descriptions provided. When descriptions are generic, duplicated, or simply echo method names, the agent cannot tell similar operations apart and may call the wrong one. The article argues this is often misdiagnosed as a model limitation, when the real issue is missing signal in the tool text. Common problems include method-name echoes, identical descriptions for related tools, and no stated boundary between operations like create and update. Writing descriptions that state what a tool does, when to use it, and when not to use it, and generating them from OpenAPI specs, is presented as a cheap way to improve reliable tool selection.