Three well-named, read-only tools with clear descriptions and complete input schemas. All tools follow verb_noun naming (search_*, dataset_info, query). Descriptions are concise (60 - 180 chars) and explain purpose and prerequisites. Input schemas include types and descriptions for all parameters. However, output schemas are completely undocumented, LLMs cannot predict response structure or plan downstream calls. Parameter descriptions lack format constraints (e.g., 'ODSQL filter' is vague; should specify syntax rules). No error handling guidance. Missing required field declarations in schemas (only dataset_info marks 'dataset_id' as required; query and search_datasets should too). No pagination metadata (total count, next_cursor) documented.

• Document output schemas for all three tools. Specify response fields, types, and structure. E.g., search_datasets returns {datasets: [{dataset_id, title, theme, record_count}], total_count}.
• Add format constraints to ODSQL parameters. E.g., 'where: ODSQL WHERE clause (e.g., year >= 2020 AND city = "Paris"). Valid operators: =, !=, >, <, >=, <=, AND, OR, IN, LIKE.'
• Mark 'dataset_id' as required in search_datasets and query schemas (add to 'required' array).
• Wrap API errors with recovery guidance. E.g., if dataset not found, return 'Dataset "xyz" not found. Call search_datasets() to find valid dataset IDs.'
• Document pagination in search_datasets and query descriptions. E.g., 'Returns up to {limit} results. Use offset to paginate. Response includes total_count for planning.'
• Add tool annotations (readOnlyHint: true) to all three tools to signal they are safe to call without user confirmation.