Server has 4 tools with complete input schemas and descriptions. Naming follows verb_noun convention (lookup_, get_, manage_). Descriptions are present (60-150 chars) but lack actionable context for LLM selection. Parameters are typed and mostly described, but lack enums for constrained values (e.g., lookup_type, level, data_type accept free-form strings). Output schemas are not documented in the visible code. Error handling and recovery guidance are absent. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.

• Add enum constraints to all parameters with known valid values: lookup_type, level, data_type, category, aggregate, action, region. Replace free-form strings with JSON Schema enum arrays.
• Document output schemas for each tool. Specify field names, types, and whether results are paginated. Include example response structure in tool documentation or code comments.
• Add tool annotations: readOnlyHint=true for lookup_agency, get_arrest_data, get_police_employment; destructiveHint=true for manage_cache. This enables agents to reason about side effects.
• Enhance descriptions with recovery guidance: 'If offense code is invalid, call get_reference_data(data_type="offenses", offense_type="arrest") to list valid codes.' Include prerequisites and common failure modes.
• Add per-parameter constraints in descriptions: 'state: Two-letter abbreviation (e.g., CA, NY). Required when level is "state" or "agency". Omit for "national" level.' This prevents invalid LLM inputs.