Two well-defined read-only tools with clear naming, descriptions, and Zod schemas. Tool names follow verb_noun convention (cityjson_read_*). Both have explicit input schemas with type definitions and parameter descriptions. Descriptions are concise (100-150 chars, within baseline) and convey purpose clearly. Parameters are minimal (1-2 per tool) and well-constrained. However, output schemas are not documented in the visible code, the tools return unstructured responses without declared field schemas. No error handling guidance or recovery patterns present. No tool annotations (readOnlyHint despite risk:READ_ONLY metadata). Limited composition: tools are isolated discovery/content retrieval with no downstream chaining requirements.

• Add explicit output schema documentation for both tools. For cityjson_read_spec_outline, document fields: chapters (array of {id, title, sections?}), metadata. For cityjson_read_spec_chapter, document: chapter_id, title, content (markdown string), length.
• Add readOnlyHint annotation to tool registration: server.tool(..., {readOnlyHint: true, ...}).
• Replace free-form chapter parameter with enum constraint. Extract valid chapter IDs from the spec index and declare them: z.enum(['metadata', 'city-objects', 'geometry-objects', ...]).
• Add error handling in handler functions. Return structured errors like {error: 'Chapter not found', availableChapters: [...], suggestion: 'Did you mean X?'} to guide LLM recovery.
• Consider adding a helper tool cityjson_list_chapters (discovery) that returns all available chapters with brief descriptions, enabling agents to self-correct before calling read_spec_chapter.