Four tools with complete input schemas and descriptions, but significant gaps in output documentation and error handling. Tool naming follows verb_noun convention (generate_*). Descriptions are present (avg 120 chars) but lack context on when to use each tool vs alternatives. No output schemas documented, LLMs cannot plan downstream calls. Error handling is minimal; no recovery guidance. Parameters are well-typed with enums for format selection, but descriptions lack constraint details (e.g., markdown syntax requirements, file size limits). Security concern: apiKey and baseURL exposed as parameters in generate_knowledge_graph, violating secret-injection pattern.

• Document output schemas for all tools. E.g., generate_mindmap returns {content: string, format: 'html'|'svg', size_bytes: number}. generate_and_save_mindmap returns {file_path: string, format: string, size_bytes: number}.
• Move apiKey and baseURL from tool parameters to server-side environment variables (OPENAI_API_KEY, OPENAI_BASE_URL). Inject at runtime in generate_knowledge_graph handler. Remove from input schema.
• Split generate_and_save_mindmap into two tools: (1) generate_mindmap_content (returns HTML/SVG string), (2) save_file (accepts content + path). Allows agents to generate without saving, or save generated content elsewhere.
• Enhance parameter descriptions with constraints: 'markdown: Valid Markdown syntax (max 50KB, nesting ≤10 levels). Use standard list markers (-, *, +) and proper indentation.' 'outputDir: Directory path. Defaults to $COGNIGRAPH_OUTPUT_DIR or ~/cognigraph/. Must be writable.'
• Implement structured error handling. Catch CLI errors and return {error: string, code: string, suggestion: string}. E.g., {error: 'Invalid Mermaid syntax', code: 'SYNTAX_ERROR', suggestion: 'Check graph definition. Common issues: missing semicolons, invalid node IDs.'}.