Server demonstrates solid tool design with clear naming (verb_noun pattern), comprehensive parameter descriptions, and structured error handling via KIOError codes. All 6 tools have descriptions and input schemas with type definitions. However, output schemas are not explicitly documented in the visible code, and some parameter descriptions lack format constraints (e.g., date_from/date_to format hints, pzp_article format examples). Tool composition is clean, each tool has a single responsibility. The INSTRUCTIONS section is exceptionally detailed and guides LLM behavior well, but this is procedural guidance rather than schema documentation. Parameter validation rules are documented in INSTRUCTIONS but not in schema descriptions themselves, which reduces discoverability.

• Document output schemas explicitly in tool definitions. For each tool, add a 'returns' section listing field names, types, and descriptions. Example: kio_search returns {results: [{signature: string, issue_date: string|null, doc_type: string, ...}], total: integer, page: integer}.
• Add JSON Schema format and pattern constraints to date parameters. Change date_from/date_to descriptions to include 'format: date' or add a regex pattern: '^\d{4}-\d{2}-\d{2}$' to enforce YYYY-MM-DD.
• Convert 'sort' parameter to an enum in the schema: {type: 'string', enum: ['rank', 'date_asc', 'date_desc']}. Remove the enum list from the description text and let the schema be the source of truth.
• Expand kio_coverage description to explain what it returns (data families, source, gaps) and when to call it (when search returns empty or question touches missing material). Add example output fields.
• Add a 'size' parameter constraint to all paginated tools: {type: 'integer', minimum: 1, maximum: 100, default: 20}. This prevents LLMs from requesting 1000 results and clarifies the hard cap.