dtc-mcp registers three tools with descriptions and partial schemas. However, critical gaps prevent a higher score: (1) execute_code lacks parameter descriptions for 'timeoutMs' beyond a bare label; (2) search_docs and read_doc have minimal input documentation; (3) no output schemas are documented anywhere in the source; (4) error handling and recovery guidance are absent; (5) no input validation guidance or constraints on code execution timeout ranges. The server is STDIO-only, which is a hard transport cap at 50, but definition quality issues independently limit it further. Naming is reasonable (execute_, search_, read_) but descriptions are terse and lack 'what to do if it fails' guidance.

• Document the output schema for each tool. For execute_code: return {success: boolean, result?: any, error?: string, executionTimeMs: number}. For search_docs: return {results: [{docId: string, title: string, excerpt: string, relevance: number}], totalCount: number}. For read_doc: return {docId: string, title: string, content: string, markdown?: string}.
• Add constraints to execute_code's timeoutMs parameter: 'Timeout in milliseconds (min 100, max 30000). Defaults to 5000. Higher values risk blocking the LLM context; lower values may timeout complex queries.'
• Expand search_docs description: 'Search Klaviyo and Shopify API documentation by keyword or phrase. Returns up to 10 results ranked by relevance. Use this to discover available API endpoints and methods before calling execute_code.'
• Expand read_doc description: 'Read the full content of a specific documentation page by ID (e.g., "klaviyo/api/campaigns" or "shopify/admin/orders"). Use this after search_docs identifies a relevant topic, or if you already know the doc path.'
• Add error recovery guidance to execute_code: 'If execution times out, try breaking the code into smaller chunks and calling execute_code multiple times. If syntax errors occur, fix and retry. If a method is not found, call search_docs to verify the SDK API.'
• Document what happens if a docId is not found in read_doc: 'Returns {error: "Document not found", suggestion?: "Did you mean: ..."}. Always call search_docs first if you are unsure of the docId.'
• Specify pagination for search_docs: 'Add optional parameters: limit (1 - 50, default 10) and offset (default 0) to support pagination. Return a next_offset or total_count to guide the LLM on further queries.'