18 tools with consistent structure but significant quality gaps. All tools have descriptions (10-200 chars, mostly adequate), but schemas are incomplete: tools with no parameters use empty `{}` objects (not proper JSON Schema), and tools with parameters use Zod types in TypeScript rather than JSON Schema format visible to clients. Parameter descriptions are present but generic. No output schemas documented. Error handling returns `{success, error}` tuples without recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Naming is verb-prefixed (CoinGecko_*) but inconsistent with MCP conventions (should be snake_case verbs like get_asset_platforms, not CoinGecko_asset_platforms). All tools are READ_ONLY, which is good for safety but not annotated in the schema.

• Rename all tools to snake_case verb_noun pattern: 'CoinGecko_asset_platforms' → 'get_asset_platforms', 'CoinGecko_coin_price' → 'get_coin_price', 'CoinGecko_search' → 'search_coins'. This aligns with MCP conventions and improves LLM discoverability.
• Document output schemas for all tools. Example for get_coin_price: 'Returns {prices: {[coinId]: {[currency]: number}}, market_caps: {...}, volumes: {...}}'. LLMs need this to plan downstream calls.
• Add readOnlyHint: true to all tool schemas since all are READ_ONLY. This enables clients to optimize caching and retry behavior.
• Enhance error messages with recovery guidance. Instead of 'Coin not found', return 'Coin "xyz" not found. Try search_coins(query="xyz") to find the correct ID, or call get_coin_list() to see all available coins.'
• For tools with array parameters (coinIds, currencies), add pagination/limit guidance in descriptions. E.g., 'Maximum 250 coins per request; use multiple calls for larger batches.' This prevents LLM-generated oversized requests.