Week 3 · Thu, Oct 1, 2026 · Tool calling, MCP servers, and Skills
Give your AI tools it can actually use
We built the loop that lets a model use real tools, from scratch, on live NYC Open Data. Then we wrapped those same tools as an MCP server so any AI app can use them, and wrote a Skill: a plain-English playbook for the whole job.
What we covered
The model never runs your code
It asks for a tool, your code runs it, the result goes back, and the loop repeats until the job is done.
A tool description is just JSON Schema
Vague descriptions make the model guess. Fix the description and the model fixes itself.
Never crash, report back
Return errors to the model as results so it can recover instead of stopping cold.
Write the tool once, use it anywhere
MCP (Model Context Protocol) lets one tool server plug into many AI apps. We used goose.
Tools are capabilities, Skills are the playbook
A Skill is a Markdown file that tells an agent how to do a whole task with the tools it has.
Try the homework
About 60 minutes. A free GitHub account. No API key: a built-in practice model works offline.
- Open the repo in Codespaces (no install) or clone it to your laptop.
- Milestone 1: run the agent loop and watch real tool calls on NYC garden and 311 data.
- Milestone 2: break a vague tool description, then fix it so the model recovers.
- Milestone 3: add a second tool so the model chains calls into a score.
- Stretch: wrap the tools as an MCP server. Bonus: give the agent a Skill.
Fellows turn in homework through Google Classroom. The Classroom link is on the last slide.
Want the full fellowship?
Classes are open to everyone. Fellows also get homework feedback in Google Classroom, a cohort Discord role, office hours, and priority seats at the hackathon.