RankPath MCP server defines 5 tools with explicit schema definitions using schemars::JsonSchema. All tools have descriptions and input parameters with type annotations. However, there are meaningful gaps: (1) Parameter descriptions are minimal and lack actionable constraints or examples. (2) No output schemas are documented, responses are JSON-serialized API payloads without field documentation. (3) Missing enum constraints for optional filter parameters (severity, status). (4) No error handling guidance, api_error() returns raw error messages without recovery hints. (5) No pagination cursors despite get_crawl_history accepting limit/offset. The tool naming is adequate (all verb-noun style), and descriptions succinctly state what each tool does, but fall short of LLM-optimized descriptions that explain WHEN to use and WHAT to expect.

• Document output schemas for all 5 tools. For example, list_projects should document that it returns { projects: Array<{id: string, name: string, url: string}> }. Use JSDoc or inline comments in CallToolResult to make fields discoverable to LLMs.
• Convert 'severity' and 'status' parameters in get_issues to enums. Change input schema from { type: 'string' } to { type: 'string', enum: ['critical', 'warning', 'info'] } for severity, and { type: 'string', enum: ['open', 'acknowledged', 'ignored'] } for status.
• Add minValue and maxValue to limit parameter in get_crawl_history: { type: 'integer', minimum: 1, maximum: 100, default: 10 }. This makes constraints machine-readable, not just documentation text.
• Enhance error responses with recovery guidance. Instead of returning raw error text, return structured errors like: { is_error: true, content: [{ type: 'text', text: 'Project not found (UUID: xyz). Call list_projects() to see available projects.' }] }. Pattern: recovery-guide.
• Add pagination metadata to get_crawl_history response. Include total_count and next_cursor (or has_more flag) so agents can iterate reliably. Return { crawls: [...], total_count: 150, next_cursor: 'offset_100' }.
• Add tool annotations to ServerCapabilities. Register all tools with { name: 'list_projects', ..., readOnlyHint: true } to make their safety profile machine-readable.
• Expand parameter descriptions to be LLM-friendly. Example: limit parameter should read: 'Maximum number of results (1 - 100, default: 10). Larger limits may increase response time.' Current description is too terse.