CryptoConduit has 9 tools with complete input schemas and descriptions, but quality is inconsistent. Naming follows verb_noun convention (get_, set_, search_) which is good. However, descriptions are often too technical and lack LLM-optimized guidance on WHEN to use each tool. Parameters are well-typed with enums for assets/windows, but several tools lack output schema documentation. Error handling is minimal, no recovery guidance or actionable error messages visible in the code. The set_auto_trade tool is particularly complex with 10 parameters, some conditionally required, but dependencies are not clearly documented. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels (WRITE vs READ_ONLY). Output formatting (text vs json) is offered but not well-integrated into schema definitions.

• Add tool annotations to all tools: mark get_* tools with readOnlyHint=true, set_* tools with destructiveHint=true. This signals to agents which calls are safe to retry.
• Document output schemas for all tools. For json format responses, specify the field names, types, and what each field represents. Example: 'Returns {market_type: string, up_ask: number, down_ask: number, spread_pct: number, ...}'
• Rewrite descriptions to be LLM-optimized (50-200 chars). Include WHEN to use: 'Get quantitative analysis (volatility, momentum, fair value) for a crypto market. Use this to assess mispricing before trading.' vs current generic description.
• Document parameter dependencies explicitly. For set_auto_trade, add to description: 'When enabled=true, all of entry_pct, min_move_pct, max_entry_price, position_size_usd are required. When enabled=false, only asset and window are needed.'
• Add error recovery guidance. Instead of returning raw errors, guide the LLM: 'Asset "xyz" not found. Valid assets: btc, eth, sol, xrp, doge, bnb. Try again with a valid asset.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry