Graphify exposes 5 read-only graph analysis tools with basic descriptions and input schemas. All tools have descriptions (10-100 chars) and parameter types, but descriptions are generic and lack LLM-optimized guidance on when/why to use each tool. Output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Error handling returns JSON errors but lacks recovery guidance. No tool annotations (readOnlyHint, etc.). Parameter descriptions are minimal; e.g., 'Search term to match against node IDs, labels, or types' lacks format/constraint details. No pagination support despite Query and Analyze potentially returning large result sets. Tool naming is clear (verb_noun pattern) but descriptions do not explain distinctions between similar tools (Query vs Explain vs Analyze).

• Document output schemas for each tool. Example for Query: 'Returns {query: string, resultCount: int, results: [{id, label, type, filePath, language, confidence, community, degree, connections: [{source, target, relationship, weight}]}]}'. This enables LLMs to extract and chain data.
• Expand tool descriptions to 100-200 chars with domain context and selection guidance. E.g., 'Query: Search nodes by ID, label, or type. Use this to find entities in the graph. Returns up to 10 matching nodes with their immediate connections. For detailed analysis of a single node, use Explain instead.'
• Add pagination to Query and Analyze: return total_count and next_cursor (or offset). Document in description: 'Results are limited to 10 by default. Use limit parameter (1-100) and pagination to retrieve larger result sets.'
• Enhance error messages with recovery hints. E.g., 'Node "xyz" not found. Try Query(searchTerm="xyz") to search for similar nodes, or Communities() to list available nodes by community.'
• Add readOnlyHint annotation to all tools (via [McpServerTool] attribute or schema metadata) to signal safety and enable agent retry logic without user confirmation.