MCP server for static security scanning of code with deterministic rules and stable finding fingerprints
Single tool 'bastion_scan' has a clear, action-oriented name and a reasonable description (87 chars). Input schema is present with 4 parameters, all typed and described. However, the description lacks critical context about what the tool returns, when to use it vs alternatives, and how to interpret results. Parameter descriptions are minimal (e.g., 'relative directory within configured project root; defaults to .' is adequate but terse). No output schema is documented, LLMs cannot predict what fields to expect from scan results. Error handling and recovery guidance are absent. The tool is read-only (good security posture), but lacks the depth expected of production-grade tools.
Read-only static security scan of code inside the configured project root.
No output schema documented. LLMs cannot predict scan result structure (findings, counts, metadata, etc.), forcing them to guess how to interpret and chain results.
Tool description (87 chars) lacks WHEN to use it, WHAT it returns, and HOW to interpret results. Should explain: 'Returns security findings with severity, location, and remediation hints. Use to audit code before deployment.'
Parameter 'rules' description is vague ('optional rule IDs to enable'). Should specify: format of rule IDs, how to discover available rules, and what happens if an invalid rule is passed.
No error handling guidance. If scan fails (e.g., invalid path, permission denied, timeout), LLM has no recovery path. Should document: retryable errors, user-fixable errors, and actionable error messages.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | <=2025-11-25 | v2 |
Parameter 'max_files' constraint is incomplete. Description says '1,10000' but should clarify: is this inclusive? What happens if scan exceeds max_files? Does it truncate, error, or paginate?