Single tool 'gis_layer_search' has a well-structured input schema with proper types, descriptions, and constraints (limit capped 1-10, skip min 0). Description is adequate (108 chars, within baseline 34-392 range) but lacks clarity on WHEN to use vs alternatives, and omits what the tool RETURNS. The schema is visible and complete, but output structure is not documented anywhere in the provided source. The tool name follows verb_noun convention ('search_') appropriately for a READ_ONLY operation. No error handling, no per-parameter type/format guidance beyond JSON Schema.

• Document the full output schema: what fields does each result object contain? (e.g., id, name, type, description, url, metadata_text, geom_bounds, distance_score?). Include example response structure.
• Expand tool description to 150-200 chars: explain that this tool combines semantic search (NLP-based) with spatial filtering (geographic intersection). State: 'Use this to find GIS layers matching a text query AND overlapping a specific location. Returns ranked layers with metadata.'
• Add enum constraint to 'type_filter' with valid layer types (if known: raster, vector, point, polygon, etc.). Or explicitly state 'Use any layer type strings from the datasource'.
• Document the 'input_point' and EPSG interaction: 'If you only have lat/lon, use EPSG 4326 (WGS84), the default. If using another projection, specify its EPSG code.'
• Add error handling in tool description: 'If no layers match, try broadening the search query or removing spatial filters. If the embedding model is unavailable, the search will fail, contact admin.'
• In the MCP server code (backend/mcp.py, not shown), ensure tool definition includes inputSchema with full type annotations and a clear response schema declaration.