Three well-named, read-only tools with clear descriptions (avg 180 chars, within baseline 194). All tools have input schemas with typed properties and required fields. However, parameter descriptions lack actionable constraints (ranges, formats, enums). Output schemas are undocumented, LLMs cannot predict response structure for chaining. Error handling is minimal (generic HTTP status errors). No tool annotations (readOnlyHint, idempotentHint). Security is solid (SSRF guard in layerUrl, no secrets exposed). Composition is clean, each tool has one responsibility, and search_datasets→query_layer→layer_info form a logical chain.

• Document output schemas for all three tools. For search_datasets, specify: {total: number|null, count: number, datasets: [{id: string, name: string, summary: string|undefined, org: string, org_id: string, record_count: number, type: string, url: string}]}. This enables LLMs to extract dataset URLs for downstream query_layer calls.
• Add enum constraints to parameters. Replace 'Max datasets (1-50, default 20)' with inputSchema properties: {limit: {type: 'number', minimum: 1, maximum: 50, default: 20}}. JSON Schema constraints are machine-parseable and prevent hallucinated values.
• Enhance error messages with recovery guidance. Instead of 'ArcGIS query: 400', return '{error: "Invalid where clause", details: "SQL syntax error near 'STATE'", suggestion: "Use format: \"FIELD = value AND FIELD2 >= value2\""}'. This guides LLM self-correction.
• Add tool annotations to the McpToolDefinition interface. Include readOnlyHint: true and idempotentHint: true for all three tools. This signals to MCP clients that results can be cached and retried safely.
• Document pagination semantics for query_layer. Clarify: 'offset is 0-indexed. To fetch records 51-100, use offset=50, limit=50. Returns {features: [...], exceededTransferLimit: boolean}.' This prevents off-by-one errors in LLM pagination logic.