TripDesk practice lab (Axel Cyber Lab, Track 14: Securing AI Agents and MCP)

Everything in this folder is fictional training material for your own computer.
Alpenrose Travel, TripDesk, every @example.test address and the marker
ACL-CANARY-7f3a are made up. Some files are deliberately unsafe so you can
watch an attack happen and then block it:

  data/inbox/msg-002.txt   an email with a hidden instruction for the AI
  web/index.html           a web page with a hidden instruction for the AI
  weather_server.py        a tool whose description hides an instruction
  server.py                the trip server with a path bug (fixed: server_fixed.py)
  booking_server.py        a server that leaks its token (fixed: booking_server_fixed.py)

Use them only in this lab. Never point them at real services or real data.

Setup (Python 3.10 or newer):
  python3 -m venv .venv && . .venv/bin/activate
  pip install mcp==2.3.0

Run the agent with the deterministic practice model:
  MOCK=1 SERVERS=server.py python agent.py "Read the inbox and summarise the messages."

Run the red-team suite (vulnerable, then hardened once you have written policy.json):
  export BOOKING_TOKEN=acl_tok_lab_example
  MOCK=1 python redteam.py server.py booking_server.py
  MOCK=1 python redteam.py server_fixed.py booking_server_fixed.py

The lessons walk you through each file step by step.
