Security audit: Arbitrary code execution with no auth (Score: 0/100)

[stevenkozeniesky02]

Hey — we ran an automated security scan of pentest-mcp using the AgentsID Scanner and wanted to share the results.

Summary

Metric	Value
Tools scanned	19
Security score	0/100 (Grade: F)
Critical findings	Arbitrary code execution, no auth
High findings	22
Medium findings	16
Risk tags	execution, mutation

Key Findings

Arbitrary Code Execution (Critical)

• runJohnTheRipper and runHashcat execute arbitrary code with no scope restriction — any input is passed directly to execution

No Authentication (High)

• No authentication mechanism — server accepts any unauthenticated connection

Missing Tool Descriptions (High/Hallucination)

• Multiple tools have no descriptions — the LLM infers behavior from the tool name alone, which leads to unpredictable execution patterns

Category Breakdown:

• permissions: F | validation: F | hallucination: F | auth: B | output: A

How to Reproduce

npx @agentsid/scanner@latest pentest-mcp

Context

We understand this is a pentest tool and many of these findings are "by design" — running JtR and hashcat is the point. But from an MCP security perspective, an agent connecting to this server with no auth and no scope restrictions means any prompt injection in the agent's context could trigger these tools. Adding basic auth and tool-level confirmation gates would significantly reduce the blast radius.

This is part of a census-scale audit of the MCP ecosystem — 15,982 servers scanned, 137,070 findings. Full research: Weaponized by Design

Full results: agentsid.dev/registry

[DMontgomery40]

Not sure y'all understand who/when/where this is run.