Security scan results: 20/100 (F) with 77 findings across 36 tools

[stevenkozeniesky02]

We're running free security audits of popular MCP servers as part of a research project on the MCP ecosystem. Your server scored 20/100 (F) with 77 findings across 36 tools.

Breakdown by category:

Category	Grade	Notes
Validation	F	36 tools with unbounded string parameters, no maxLength or pattern constraints
Hallucination	F	Multiple tools use vague descriptions that cause LLMs to over-interpret scope
Auth	B	No authentication-related tools detected
Permissions	A	No destructive/execution tool names

Top findings:

• git_discard_changes, git_unstage, git_amend, git_stage all use vague action words ("change") without specifying what they operate on. LLMs interpret this as the broadest possible scope.
• 36 tools exposed with no per-tool permission boundaries. With tools like git_discard_changes and git_amend available, a prompt-injected agent could destructively modify a repo.
• No input schema constraints on most tools. String parameters accept arbitrary input with no validation.

Most of these are structural patterns common across the MCP ecosystem (72% of the 15,982 servers we scanned scored F). Happy to walk through specific fixes for any of these.

Full results: https://agentsid.dev/registry
Scanner is open source: npx @agentsid/scanner