Guard MCP has well-structured schemas with Zod validation and clear parameter constraints (enums, min/max bounds). Tool names follow verb_noun convention (verify_code, verify_changed_files, generate_contract_rules). However, descriptions are terse (10-50 chars), lacking context on WHEN to use each tool and what distinguishes them. Parameter descriptions exist but are minimal. Output schema is documented via TypeScript types but not exposed in tool registration. Error handling returns structured responses but lacks recovery guidance (e.g., 'spec not found' doesn't suggest how to generate one). No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.

• Expand tool descriptions to 100-200 characters. For verify_code: 'Verify a single code file against contract rules. Use this for targeted checks on one file. For batch verification of all changed files, use verify_changed_files. Requires a contract spec (auto-generated if missing).'
• Add tool annotations to tool registration: readOnlyHint=true for verify_* tools, destructiveHint=true for generate_contract_rules (writes to disk).
• Document output schema in tool descriptions or as a structured field. E.g., 'Returns: {ok: boolean, violation_count: number, violations: [{rule_id, line, description, fix_suggestion}]}'
• Enhance error messages with recovery paths. E.g., 'Spec not found at <path>. Auto-generate one with generate_contract_rules, or provide spec_path parameter.'
• Add parameter guidance: 'profile: auto (detect from workspace), frontend-vue3-ts-antdv (Vue 3 + TypeScript + Ant Design), generic-frontend (fallback). Use auto unless you have a specific framework.'