ReadyPermit defines 5 tools with clear action verbs (analyze_, check_, compare_, what_can_i_build). All tools have descriptions (194 - 280 chars, within baseline 34 - 392 range) and input schemas with Zod validation. However, output schemas are NOT documented, responses return unstructured text blobs rather than typed objects. Parameter descriptions are present but generic (e.g., 'Full U.S. property address' repeated 4 times). No error handling guidance, no recovery paths, and no pagination despite tools that could return large result sets. Tool composition is reasonable (each does one thing), but responses lack the structured fields needed for agent chaining.

• Document output schemas for all tools. Return structured objects with typed fields: {property: {address, lat, lng, city, state, zip}, zoning: {designation, constraints}, buildability: {score, grade}, risks: {flood_zone, epa_sites, wildfire}, report_url}. This enables agent reasoning and chaining.
• Add error handling with recovery guidance. If geocoding fails, return {error: 'Geocoding failed: MAPBOX_TOKEN not set. Set the environment variable and retry.', fallback_address: <input>}. Categorize errors as retryable vs fatal.
• Enrich parameter descriptions with format constraints and examples. E.g., 'Full U.S. property address (e.g., 123 Main St, Austin, TX 78701). Required: street, city, state, ZIP. Max 200 chars.'
• For compare_parcels, return an array of property objects with per-property scores and risk levels, not markdown. Enable programmatic comparison: [{address, zoning, buildability_score, flood_zone, adu_eligible}, ...].
• Include property IDs or normalized addresses in responses so agents can chain tools. E.g., analyze_property returns a property_id that check_adu_eligibility accepts, enabling multi-step workflows without re-geocoding.