Add MCPGoat — a deliberately vulnerable Model Context Protocol (MCP) target - #276
Add MCPGoat — a deliberately vulnerable Model Context Protocol (MCP) target#276SabyasachiDhal wants to merge 1 commit into
Conversation
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
You clearly have not searched for MCP on https://vwad.owasp.org/ 😁 |
| "TypeScript", | ||
| "Node.js", | ||
| "Docker" |
There was a problem hiding this comment.
Seem like this should include: MCP, LLM, AI
|
Ha, you're right — I searched the repo's MCPGoat's angle is depth: 26 challenges, each hardened across Easy / Moderate / Difficult plus a Secure reference level (78 flags), covering MCP-specific attacks (tool poisoning, rug-pulls, sampling abuse) alongside the classic web bugs, with a victim-agent harness that shows a real LLM agent being exploited. Happy to tweak the wording or category so it sits well alongside the others — thanks for taking a look! |
|
If you've got a logo you should include it as well |
|
Ugh, just noticed you're on the old repo. Please join us at our new home: https://vwad.owasp.org |
Adds MCPGoat to the directory — a deliberately vulnerable Model Context Protocol (MCP) server for hands-on penetration-testing practice.
As AI agents adopt MCP, MCP servers are a fresh, under-tested attack surface. MCPGoat implements 26 challenges spanning MCP-specific attacks (tool poisoning, tool shadowing, rug-pulls, indirect prompt injection, sampling abuse) and classic web bugs (RCE, SSRF, SQLi, path traversal), each at Easy / Moderate / Difficult plus a Secure reference level — 78 capture-the-flag flags — with a victim-agent harness that shows a real LLM agent being exploited.
Conformance to CONTRIBUTING.md / schema.json:
_data/collection.json, tab-indented.schema.json(all keys, enums, and required fields).name(lands between "Marathon" and "Metasploitable 2").ctf+single-player;collection=offline+container(ships a Dockerfile / docker compose).🤖 Generated with Claude Code