This HTTP-based FastAPI MCP server exposes 8 read-only tools for a news aggregation platform. All tools have descriptions and input schemas are visible in the code. However, the schemas lack depth, most parameters are missing detailed constraints, enums, and type specifications beyond basic types. Descriptions are present but generic (avg ~80 chars), falling short of the LLM-optimized 50-200 char range. Parameter descriptions exist but are minimal ('Search query', 'Filter by topic'). Output schemas are not documented anywhere in the source. Error handling and recovery guidance are absent. The server is composition-sound (each tool has one responsibility) but lacks polish expected of production-grade agent toolkits.

• Document output schema for every tool. Example: 'list_articles returns {articles: [{id, title, source, date, is_ai_related, category, url}], total: int, next_offset: int|null}'. This enables LLMs to chain calls and extract required IDs.
• Add enums to source_type and exclude_source_type parameters in list_articles: "enum": ["rss", "newsletter", "github", "crawler"] in the schema.
• Expand tool descriptions to 80-150 characters, including WHEN to call each tool. Example for semantic_search: 'Find articles by meaning rather than keywords. Uses vector embeddings; falls back to keyword search if embedding fails. Better than list_articles for conceptual queries like "AI safety concerns".'
• Add parameter constraints to limit and offset: maximum=100 for limit, minimum=0 for offset. Document in parameter descriptions: 'Result limit (1-100, default 20).'
• Document date format requirements. Parameters from_date, to_date, and target_date declare format: date, but add description clarification: 'ISO 8601 date string (YYYY-MM-DD). Example: 2024-12-15.'
• Add error recovery guidance to semantic_search description: 'If semantic search fails (embedding service down), automatically retries with keyword ILIKE search. LLM will receive results from fallback method.'
• Remove health from agent-facing tools or mark it as a health-check-only endpoint, not a user-facing capability. Health checks are infrastructure, not agent operations.
• For list_articles and list_summaries, document pagination behavior: 'Returns up to limit results. Use offset to fetch subsequent pages. Repeat with offset += limit until fewer than limit results returned.'