This MCP server has significant definition quality gaps. While it registers 9 tools with names that follow verb conventions, most descriptions lack depth and parameter documentation is inconsistent. Tool #4 (um_build_pipeline_code) has extensive parameters but many lack descriptions or type constraints. Tool #1 (um_list_urban_layers) has empty input schema. No output schemas are documented, making it unclear what structures agents should expect. Error handling is minimal, no guidance on recovery steps or categorization of errors (retryable, user-fixable, fatal). The descriptions average 10-50 characters, well below the 194-char baseline for A+ tools, leaving LLMs unable to make confident selection decisions. The tool definitions are directly visible in the code (good), but the lack of schema completeness and sparse parameter documentation keeps scores conservative.

• Add output schemas to all tools. Document return types in tool definitions: e.g., um_list_urban_layers returns list[str] of layer type names; um_hf_dataset_schema returns {schema: dict, samples: list[dict], column_mappings: dict}. Enable agents to plan chaining and validate data.
• Expand discovery tool descriptions from 30 chars to 100 - 150 chars. Answer: What does it return? When should I call it? What's the next step? E.g., 'List available urban layer types (e.g., grid, hexbin, census_tract). Call this first to see options before building a pipeline. Returns list of type IDs to pass to um_build_pipeline_code.'
• Add missing parameter descriptions to um_build_pipeline_code. Document each of: number_of_rows (limit rows loaded; if None, load all), streaming (bool: use HF streaming mode), include_imputer (bool: add geo-imputation for missing coords), address_column (column name for AddressGeoImputer), include_filter (bool: add bounding box filtering), csv_path (output file path; if None, no CSV written), visualise_columns (list of column names to visualize; if None, visualize all).
• Document parameter dependencies in um_build_pipeline_code: 'Either pass (longitude_column + latitude_column) for point geometry, OR geometry_column for WKT. Not both.' 'For aggregations other than count, values_from is required.' 'address_column is only used if include_imputer=true.'
• Replace visualiser_style parameter with constrained options. Provide um_visualiser_style_schema tool (already exists) that returns valid keys/types for each visualizer, or define an enum of preset style names and let um_build_pipeline_code accept visualiser_preset instead of freeform dict.