Vectimus exposes 6 tools via HTTP, but lacks critical definition quality. Tool descriptions are present but minimal (10-80 chars, well below the 50-200 char baseline for LLM-optimized descriptions). Input schemas are either missing entirely or trivial. The 'evaluate' tool accepts a generic 'body' parameter with no type constraints or field documentation. Health/readiness probes are operational tools, not agent-facing capabilities. No parameter descriptions, no output schemas documented, no error recovery guidance. This server is a governance/policy engine, not an agent tool interface, the tools exposed do not follow agentic tool patterns.

• Expand tool descriptions to 50 - 200 characters. Example: 'evaluate' → 'Evaluate a tool action against Cedar policies. Returns allow/deny decision with reason. Use before executing any agent action.'
• Document the 'body' parameter schema with explicit field types and descriptions. Specify what fields the Cedar policy engine expects (e.g., principal, action, resource, context).
• Add output schemas for all tools. Document the structure of 'evaluate' responses (decision, reason, policy_id), 'list_policies' responses (policy array with name/version/status), and 'health' responses (uptime, policy_count, status).
• Add parameter descriptions for all tools. Even empty-input tools like 'list_policies' should document what they return and when to call them.
• Remove or repurpose health/readiness probes. If they must be exposed, clearly mark them as infrastructure tools, not agent-facing capabilities. Consider moving them to a separate /health endpoint outside the MCP tool interface.