Tool Descriptions - Tools Against Typed APIs, with MCP as the Wrapper.

Follow me on Linkedin:   / aziz-ketata-08ba20158   After this episode the viewer can look at two tool cards and predict, from the description text alone, whether the agent's choice between them is reliable or a coin flip. The description is prompt text the model reads at runtime, written at design time with no chance to correct it in the moment, and overlap between descriptions and names, not the number of tools, is what makes selection fail. One rewritten sentence, usually the one that says what the tool does not do, changes which tool fires.