Skip to content
Back to search

rta-mcp-ctf

Self-hosted mini CTF on attacking AI agents and MCP servers. Docker Compose, CTFd, ten simulated challenges.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/torjan0--rta-mcp-ctf/badge.svg)](/listings/torjan0--rta-mcp-ctf)

HTML

<a href="/listings/torjan0--rta-mcp-ctf"><img src="/api/listings/torjan0--rta-mcp-ctf/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 58 lines
  1. # MCP Agent CTF on your machine
  2. Play ten browser challenges about how AI agents handle untrusted tools, messages,
  3. forms and tasks. The Forensics challenges use local artifacts. In the Attack
  4. challenges you play an attacker against a simulated organization. All companies,
  5. accounts, secrets, mail, packages and tools are fictional and stay inside the lab.
  6. ## Start
  7. Install Docker Engine or Docker Desktop with Compose. Allow about 2 GB of free
  8. disk for the images and state. From this directory run:
  9. ```sh
  10. docker compose up --build
  11. ```
  12. Open the [scoreboard](http://127.0.0.1:8000) and register a player account with
  13. a password of at least eight characters. Use the same name or email and password
  14. to sign in at the [lab](http://127.0.0.1:8100). Submit flags on the scoreboard.
  15. Both services listen on your computer's loopback address. If Docker runs on a
  16. remote host, use a secure tunnel rather than exposing these ports to the internet.
  17. ## Pick an agent backend
  18. The stack chooses a backend each time it starts:
  19. 1. Set `ANTHROPIC_API_KEY` in your shell for Claude Haiku 4.5 through the Anthropic API.
  20. 2. Or set `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY` for an account with
  21. Bedrock access to Claude Haiku 4.5. `AWS_SESSION_TOKEN` is passed through for
  22. temporary credentials. Set `AWS_DEFAULT_REGION` if needed.
  23. 3. With neither, the lab uses a scripted stand-in. The hub clearly labels this
  24. practice mode. It is useful for exploring the interfaces, but its behavior
  25. does not predict a real model's response to an attack.
  26. For example:
  27. ```sh
  28. ANTHROPIC_API_KEY=your_key docker compose up --build
  29. ```
  30. Each real model run costs a small amount. At the app's default accounting rates
  31. for Haiku 4.5, input is about $1 and output about $5 per million tokens; a
  32. short run is usually a fraction of a cent to a few cents. Your provider bills
  33. your own account. The lab also limits runs per player and total estimated spend.
  34. Provider prices can change, so check your provider before running many attempts.
  35. The A2 and A3 repositories are simulated in memory. Write issues directly on
  36. their lab pages. No GitHub account or token is needed.
  37. ## Reset
  38. `docker compose down` stops the stack and keeps accounts, scores and flags.
  39. To start a fresh game, run `docker compose down -v`, then `docker compose up --build`.
  40. The `-v` option permanently deletes this stack's accounts, scores and flags.
  41. All challenge effects are simulated. Do not put real credentials or personal
  42. information into the challenge forms.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.