ATTOM MCP has 8 tools with complete input schemas and descriptions, but critical gaps limit production readiness. All tools follow verb_noun naming (attom_*) and have 50-150 char descriptions meeting baseline length (p10=34, p90=392). However, output schemas are completely undocumented, LLMs cannot plan downstream calls or extract structured data. Parameters lack type constraints (all strings, no enums for known-value fields like propertyType, interval). The _apiKey parameter violates secret-injection pattern by exposing credentials as a tool parameter. Error handling is minimal, no recovery guidance, no categorization. No tool annotations (readOnlyHint, idempotentHint). Descriptions are functional but lack dependency hints and WHEN-to-use guidance. Per-tool average: 52/100.

• Document output schema for each tool. Example for attom_property_detail: 'Returns {address, lotSize, squareFeet, bedrooms, bathrooms, yearBuilt, constructionType, heating, cooling, ...}' with field types and descriptions.
• Move _apiKey to server-side injection. Accept ATTOM_API_KEY environment variable at server startup. Remove _apiKey from all tool parameters. Update extractKey() to read from process.env instead.
• Add enum constraints: propertyType: enum [SFR, CONDO, APARTMENT, MULTI_FAMILY, ...]; interval: enum [monthly, quarterly, yearly]. Prevents hallucinated invalid values.
• Wrap attomGet() errors with recovery guidance. Example: 'Address not found. Verify spelling or try attom_property_search() with postal code. Got: {status, message, suggestion}'.
• Add tool annotations to inputSchema. Mark all tools with readOnlyHint: true and idempotentHint: true. Helps agents reason about safety and retry behavior.