Why use MCP instead of an agent skill for something like this when MCP is typically context inefficient?

pertymcpert 2 hours ago | parent | next [-]

MCP is fine if your tool definition is small. If it's something like a sub-agent harness which is used very often, then in fact it's probably more context efficient because the tools are already loaded in context and the model doesn't have to spend a few turns deciding to load the skill, thinking about it and then invoking another tool/script to invoke the subagent.

▲

wahnfrieden 3 hours ago | parent | prev [-]

Models haven't been trained enough on using skills yet, so they typically ignore them

▲

andai 3 hours ago | parent [-]

Is that true? I had tool use working with GPT-4 in 2023, before function calling or structured outputs were even a thing. My tool instructions were only half a page though. Maybe the long prompts are causing problems?

	▲	pertymcpert 2 hours ago \| parent [-]
		They're talking about "skills" which are not the same thing as tools. Most models haven't been trained on the open SKILL spec, and therefore aren't tuned to invoke them reliable when the need occurs.