Everyone's building AI agents with MCP right now, but almost nobody is testing what happens when things crash halfway through a tool call.

Spoiler Alert: Your agent loses everything and you have no idea where it stopped.

In this workshop, we'll build an AI agent from scratch that talks to tools via MCP. First we'll build it the naive way, break it on purpose, and see how bad it gets.

Then we'll rebuild it by wrapping MCP tools in durable workflows so your agent survives crashes, restarts, and network failures as if nothing happened.

No slides. Just code. You break things, you fix things, you leave with something real.

Tech Stack needed:

- Python 3,12+
- Temporal
- AWS Strands Agents or Langchain
- Docker
- mem0

Rest we can do live, even if you're a complete beginner,