AI Agents and mcp security risks
Credit: This CTF challenge comes from Joff Thyer’s webcast on October 30th, 2025.
For this lab you need a running instance of Damn Vulnerable MCP Server (dvmcp), an MCP client that can use tools and resources (such as Claude Desktop or Goose — see the client compatibility list), and MCP Inspector for some challenges.
Run dvmcp (requires git and Docker):
git clone https://github.com/harishsg993010/damn-vulnerable-MCP-server cd damn-vulnerable-MCP-server docker build -t dvmcp . docker run -p 9001-9010:9001-9010 dvmcp # MCP Inspector npm install -g @modelcontextprotocol/inspector
Each challenge is served over SSE at http://127.0.0.1:90XX/sse — Challenge 1 is port 9001, Challenge 2 is 9002, and so on up to Challenge 10 on 9010. For Claude Desktop, install mcp-remote (npm install -g mcp-remote) and add each endpoint to claude_desktop_config.json; for Goose, add each as a custom SSE extension.
Challenges
Sign in to take this quiz
Your answers are saved against your account, so you can come back to them and pick up where you left off.
Sign inRelated Courses by Antisyphon Training
Check out these courses related to the topic!

Suspicious Cloud Login
A SOC analyst notices a successful login to a cloud admin account outside of normal business hours.
Your goal is to identify what makes this login suspicious.
Observation
Cloud audit logs show: - User: admin@corp.local - Login time:
03:17 AM - Country: Russia - MFA: Not Used
All previous logins for this user came from the local office network during work hours.