CourseModel Context Protocol: The Complete Course · Module 10: Design Patterns and Agentic Systems · part 72 of 77
Part 72 · Module 10: Design Patterns and Agentic Systems

Lesson 4: Measuring whether a tool change helped

38 min read·13 Oct 2026
Module 10 · Lesson 4 · About 50 minutes · Needs: Module 3 Lesson 2 (selection accuracy, Wilson interval, sign test); Module 5 Lesson 2 (the packaged server); Module 6 Lesson 3 (the host, the calendar); Lesson 1 (the stand-in, check_availability)

The situation

Wednesday 7 October 2026, week 5. Tomas has seen the assistant create a note when a search would have done, so he widens the search_notes description: "Use this first for any question about the lab ... and before you create or save a new note." It reads well. Is it better?

One of course-guide/06's 16 evaluation cases runs through the lesson: "Are the EEG caps free the week of 12 October?" The right first tool is calendar__check_availability; the right answer is no, because Priya's pilot recordings (booking b4) take the caps on 13 October.

Are the EEG caps free the week of 12 October? stand-in picks a tool check_availability eeg-caps, 2026-10-12 "free": false b4 blocks it
The running example: the stand-in (amber: whatever chooses) calls the calendar's check_availability (blue: your system) and gets the right answer, not free (green).

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.