This MCP server exposes a single tool, optimize_keywords, with significant quality gaps. The tool has a reasonable description (85 chars), but the input schema lacks parameter descriptions, a critical flaw. The schema defines target_keywords as list[str] and page_copy as string, but neither parameter has a description field explaining what values are expected, constraints, or use cases. The output schema is entirely undocumented; there is no specification of the returned fields (optimized_copy, semrush_keyword_overlap_score, status) or their types. The implementation logic is simplistic (substring matching for overlap calculation) and does not align with real Semrush AIO optimization practices. No error handling is present, the tool will silently fail or return misleading results if target_keywords is empty or page_copy is null. No input validation guards against edge cases. The tool design does not follow the verb_noun naming pattern strongly (optimize_keywords is acceptable but generic), and it lacks guidance on when to use it or what constitutes success.

• Add description fields to both input parameters. Example: target_keywords: 'Array of 1-10 primary and secondary keywords to optimize for in the page copy. Keyword phrases are case-insensitive. Example: ["AI SEO", "search optimization"].' page_copy: 'The full page content (at least 50 characters) to analyze and optimize. Plain text or HTML accepted. HTML tags are stripped before analysis.'
• Document the output schema explicitly. Add a description to the tool that specifies: 'Returns an object with: optimized_copy (string): The input page_copy prefixed with a keyword focus statement. semrush_keyword_overlap_score (float, 0.0 - 1.0): Fraction of target_keywords found in page_copy (0.0 = no matches, 1.0 = all matched). status (string): Either "fully_optimized" (overlap >= 0.8) or "needs_semantic_update" (overlap < 0.8).'
• Implement input validation: Reject target_keywords if empty or longer than 10 items; reject page_copy if less than 50 chars or null. Return structured errors: 'target_keywords must contain 1-10 items. Received: 0.' This guides the LLM to retry with valid inputs.
• Replace substring matching with a proper semantic overlap algorithm. Consider: (a) tokenization and stopword removal, (b) morphological stemming or lemmatization, (c) optional synonym expansion via a lookup table. Current logic incorrectly counts 'AI' as a match for 'AI SEO' (partial match) and misses 'search engine optimization' for 'SEO' (synonym).
• Expand the tool description to 150 - 250 characters explaining: This tool identifies how well a page addresses target keywords using semantic overlap analysis. Use it to measure keyword integration, find gaps before content updates, or validate that a page matches the intended keyword strategy. For semantic improvements, expand coverage by adding synonym phrases or rephrasing to naturally include missing keywords.