Two tools with clear verb-noun naming (search_datasets, get_dataset) and reasonable descriptions (194 - 250 chars, within baseline 34 - 392). Input schemas present with type definitions and parameter descriptions. However, output schemas are completely undocumented, LLMs cannot predict response structure for planning. Parameter descriptions lack constraints (e.g., 'limit' has no min/max enforcement stated). No error handling guidance. Missing tool annotations (readOnlyHint, idempotentHint). Descriptions are functional but generic; they do not explain WHEN to use each tool vs. the other or what downstream operations become possible.

• Add explicit output schema documentation to both tools. Define the structure of {total, results} for search_datasets and {id, name, title, ...} for get_dataset. Include field types and descriptions so LLMs know what to expect.
• Add JSON Schema minumum and maximum constraints to the 'limit' parameter: {"type": "number", "minimum": 1, "maximum": 50, "default": 10}. This makes validation machine-readable and prevents invalid LLM inputs.
• Wrap error responses with actionable recovery hints. E.g., catch errors in callTool and return {error: 'Dataset not found. Try search_datasets(query="...") to find available datasets.', retryable: false}.
• Add tool annotations to the McpToolDefinition interface: readOnlyHint: true, idempotentHint: true for both tools. This signals to agents that these calls are safe to retry and do not modify state.
• Enhance descriptions with workflow guidance. E.g., search_datasets: '...Returns dataset IDs and summaries. Use get_dataset(id) to fetch full metadata and download links.' This clarifies the two-step pattern.