Building MCP Profile Hub part 3, when more reasoning makes the agent worse
This article explores an unexpected result from evaluating Contentstack’s MCP Profile Hub, where increasing a model’s reasoning effort made straightforward CMS agent tasks slower and less reliable. While enriching tool definitions with account-specific context clearly reduced tool calls, latency, and reasoning tokens, turning up the reasoning setting caused the model to distrust simple, correct tool responses and invent extra work. A control task listing global fields showed high-effort runs repeatedly ignoring a perfect API answer and exhausting their turn budget. The piece argues that reasoning effort must match the shape of the work, that bounded API operations often benefit from lower reasoning, and that agent evaluations should measure behaviour—calls, latency, traces, and early stopping—rather than abstract “intelligence” alone.
- ai
- api
- CMS
