# MCPhantom

A local web dashboard for reconnaissance and security testing of MCP endpoints, discovering exposed resources, tools and prompts and probing them for vulnerabilities such as injection, SSRF and path traversal.

MCPhantom points at any MCP URL, discovers its exposed resources, resource templates, tools and prompts, and lets a user interact with them from a security console interface. It runs locally with no cloud dependency and no setup beyond Python.

It is built as an extensible auditing framework rather than a turnkey scanner: it ships representative probes for classes such as SQL injection, command injection, SSRF, path traversal, IDOR and information disclosure, with findings scored from response signals like errors, data leaks and reflected output. Coverage is not exhaustive compared with dedicated fuzzing tools, and the project expects targeted, manual auditing alongside the automated probes.

It is open source so payload libraries, plugins and classifiers can be extended, integrated into CI/CD, or tuned for a specific target. An included AutoPwn feature is marked as a work in progress and intended only for CTFs or local labs.

Website: https://github.com/Dragkob/MCPhantom