Week 16: Exercises & Checklist
Graded exercises
Easy: Run
notebooks/01-mcp-server-lab.ipynbend-to-end (no API key needed). Copy the tool-schema JSON fortrack_shipment(as returned bylist_tools()) into the Week 16 sheet of the Excel tracker.Standard: Add a fourth tool to the MCP server,
get_bol(bol_id), backed byzoro.data.bol_samples(). Re-run the client connection and call the new tool through the protocol; confirm its result matches a direct call todata.bol_samples().Stretch: In
notebooks/02-multiagent-triage-team.ipynb, add a fourth specialist (billing) and its routing rule, re-run the A/B on the same 10 tickets, and rewrite the architecture justification using the new accuracy/latency/cost numbers.Portfolio: Commit the MCP server, the triage team, and the A/B report (
notebooks/02output). The report, one table plus one paragraph of justification, is a hiring-manager-readable artifact (tracked incurriculum/projects/README.md).
Hints
- Easy: Run notebook 01 top to bottom; the
ROUND_TRIP_TOOLSnumber should be3when themcpSDK is present. Copy thetrack_shipmentschema exactly as cell [10] prints it. - Standard: Add
get_bol(bol_id)to the server and toMANUAL_SCHEMAS;bol_idlooks likeZRL-10000. Reconnect the client (a newstdio_clientsession) solist_tools()picks up the fourth tool, then assert its result equals a directdata.bol_samples()call. - Stretch: Add the
billingspecialist by extendingsupervisor_classify,CAT_TO_ROUTE, and theadd_conditional_edgesmap together; re-run the same 10 tickets and rewrite the justification using the newACCURACY_DELTA. - Portfolio: The A/B report is the artifact, so ship the table and the justification paragraph; a reviewer should see the numbers and your decision in the same breath.
Checklist (mirrors manifest.json + Excel tracker)
- [ ] Mon: Study orchestration patterns and MCP (knowledge-base/10 + knowledge-base/11).
- [ ] Tue: Build an MCP server exposing ZoroLogistics tools; test with an MCP inspector.
- [ ] Wed: Build the multi-agent triage team (supervisor + 3 specialists).
- [ ] Thu: A/B test: multi-agent vs single agent on the same 10 tasks; measure quality and cost.
- [ ] Fri: Use case: publish the A/B report; justify your architecture decision.
- [ ] Sat: Take the Week 16 quiz (quiz.md), pass with 8/10; record the score in Notes.
- [ ] Milestone: Update the Excel tracker; commit the team and the MCP server.