Something I would mention, specifically with the example you proposed, is that despite the great standardization provided by MCP, the parameter usage and tool utilization by LLMs isn't deterministic. e.g. the model may not always leverage the GitHub search API optimally rather than defaulting to a Google-like search approach
Something I would mention, specifically with the example you proposed, is that despite the great standardization provided by MCP, the parameter usage and tool utilization by LLMs isn't deterministic. e.g. the model may not always leverage the GitHub search API optimally rather than defaulting to a Google-like search approach