Initech's support assistant calls tools, and the team wants to plug in three more MCP servers. Your call: what has to be true about those tools before you ship?
Key idea: A model chooses a tool only from its name and description, and every tool you connect is sent with every request unless your stack loads tools on demand. Clear descriptions get the right tool called; extra tools cost tokens even when they don't help.
Finish a run to see what a PM takes away from this lesson.
Right tool vs. tokens per request
Run a setup to plot it here
Things to try
- Run Clear with 4 tools. Does the model ever pick the wrong one?
- Switch to Vague. The tools do exactly the same jobs. Which ones does the model now pile onto, and why?
- Run Clear with 20 tools. Accuracy holds, but look at the input tokens per request. What are you paying for?
- Run Vague with 20 tools. What happens when names are vague and there are look-alikes to choose from?