Israeli Real Estate MCP Server providing access to Israeli government real estate data through the Govmap API
The Nadlan-MCP server demonstrates a solid foundation with 10 clearly named tools serving a specific domain (Israeli real estate). All tools have descriptions and input schemas with typed parameters and descriptions. However, there are notable gaps in output schema documentation, parameter constraint completeness, and error handling guidance. The naming is clear and action-oriented (get_, search_, find_, analyze_, compare_), and most parameter descriptions are present. The server follows fastmcp conventions and includes reasonable defaults. Key weaknesses: (1) Output schemas are not documented in tool definitions, LLMs cannot predict response structures; (2) Several parameters lack explicit constraints (e.g., years_back, radius_meters have defaults but no min/max bounds stated in descriptions); (3) No evidence of error handling patterns or recovery guidance; (4) Parameter relationships (e.g., deal_type filter behavior) could be clearer. The strip_bloat_fields function (visible in source) shows awareness of token efficiency, which is good practice, but the tool definitions themselves do not describe what fields are returned or how they are shaped.
Analyze market trends for a specific address including price statistics, market activity, and investment potential.
Search and autocomplete Israeli addresses.
Compare multiple addresses to identify the best investment opportunities based on market metrics and statistics.
Find recent real estate deals for a specific address with filtering and analysis options.
Get comprehensive statistics for real estate deals in a specific area.
Get polygon metadata within a radius of coordinates. NOTE: This returns polygon/area metadata, NOT individual deal transactions! Use find_recent_deals_for_address() to get actual deals.
Output schemas not documented in tool definitions. LLMs cannot predict response structure or plan downstream operations. The source shows strip_bloat_fields() removing fields, but tool definitions do not state what the final response looks like (e.g., fields returned by get_street_deals, their types, and nesting).
Parameter constraints incomplete. years_back and radius_meters have defaults but descriptions do not state min/max bounds or valid ranges. E.g., 'How many years back to search' without specifying the valid range (1 - 50? 1 - 10?). This invites out-of-range requests from LLMs.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 62 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 32 | - | v1 |
Get market activity and velocity metrics for a specific area.
Get real estate deals for a specific neighborhood polygon.
Get real estate deals for a specific street polygon.
Find comparable properties and estimate property value based on recent similar deals.
No error handling or recovery guidance in tool descriptions. No indication of what errors might occur (e.g., address not found, no deals in radius, invalid polygon_id) or what the LLM should do next. Agents have no fallback strategy when a call fails.
Parameter relationships not clearly documented. The deal_type filter uses integer codes (1=first hand/new, 2=second hand/used) rather than enums. Descriptions mention these mappings inline, but LLMs often fail to parse enum rules from prose, formal enums are self-documenting and prevent hallucinated values.
Pagination not visible in tool definitions. Tools like get_street_deals and get_neighborhood_deals accept a limit parameter but no offset/cursor/page mechanism is documented. Large result sets can exhaust context; agents need pagination to handle unbounded queries gracefully.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only and idempotent per the Risk field, but this is not declared in the tool definitions in a machine-parseable way. MCP 2026-07 supports tool annotations; using them would improve LLM planning.