Two tools registered via MCP SDK with moderate quality. Both have adequate descriptions (>100 chars) explaining use cases and supported chains. Input schemas are present with type definitions and examples. However, several quality gaps limit the score: (1) parameter descriptions lack constraint details (range, regex, character limits); (2) no explicit output schema documentation visible in the tool definitions; (3) missing error handling guidance and recovery instructions; (4) no tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite both tools being read-only; (5) parameter relationships and dependencies not documented. The tool names follow verb_noun convention (analyze_, scan_) which is good, but parameter validation appears to occur at runtime without exposure in the schema. Both tools are READ_ONLY, but this is not declared in the schema using tool annotations.

• Add explicit output schema documentation for both tools. Include return field names, types, and descriptions (e.g., 'risk_score: number; absorption_risk: string; upcoming_events: Array<{timestamp, severity, impact}>'). This enables LLMs to extract and chain results.
• Add readOnlyHint annotation to both tool schemas: {..., "readOnlyHint": true} to signal that these tools are safe to retry and have no side effects.
• For 'analyze_token_supply_risk', document the parameter relationship: 'Provide either token_symbol (for registry-backed analysis) OR both token_address and chain (for on-chain analysis). Do not provide both simultaneously.'
• For 'timeframe_days' and 'days_ahead', add range constraints: 'Must be 1 - 365; default 30. Values >365 may incur high RPC costs.'
• For 'limit', add upper bound: 'Maximum 100; default 50. Larger limits may timeout.'
• Expand 'simulation_params' documentation: 'Optional object with keys: price_shock_pct (number, 1-100), volume_shock_pct (number, 1-100), unlock_multiplier (number, 1.0-10.0). Run market shock simulation if provided; otherwise defaults to static analysis.'
• Document error cases and recovery paths: 'If token_symbol not found in registry, try token_address + chain. If RPC timeout, retry with longer timeframe_days. If chain unsupported, supported chains are: ethereum, arbitrum, bsc, base.'
• Add pagination guidance to 'scan_upcoming_unlocks': 'Results are sorted by severity (critical → low). Use limit to control batch size. Consider multiple calls with min_severity=critical first, then medium, then low for progressive discovery.'

Inferred effective spec: 2025-06-18+.

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