Four tools with clear verb-noun naming (search_*, get_*, lookup_*, compare_*) and detailed descriptions (150-250 chars each). All tools have input schemas with typed parameters and descriptions. However, output schemas are not formally documented, responses are inferred from code. Parameter descriptions are strong but lack explicit constraints (e.g., 'days: 1-16' stated in description but not as JSON Schema min/max). No error handling guidance for common failures (invalid spot ID, API rate limits, network timeouts). Tool descriptions include helpful dependency hints ('call search_spots first'), which is excellent. Missing: per-parameter type validation rules, enum constraints, and structured error recovery paths.

• Add JSON Schema definitions for all tool outputs. Document the structure of forecast objects, search results, and metadata responses so LLMs know what fields are available.
• Express parameter constraints in JSON Schema: add 'minimum', 'maximum', 'default', and 'enum' fields. Move 'days: 1-16' from description into schema minItems/maxItems.
• Enhance error responses with recovery guidance. E.g., lookup_spot should return: {"error": "Invalid spot ID. Call search_spots(query) first to find the ID, or provide a Surfline URL.", "input": text}
• Remove hardcoded token from source. Delete _DEFAULT_TOKEN and require SURFLINE_TOKEN env var. Document token rotation and fallback behavior.
• Add error classification to compare_spots. Return {"spot_id": sid, "error": str(exc), "retryable": True/False} so the LLM knows whether to retry or escalate.