GraphGraph has clear naming conventions (verb_noun pattern) and comprehensive parameter schemas with type definitions. However, tool descriptions are inconsistent in depth and quality. Several tools lack descriptions entirely (get_neighborhood, update_node, update_edge, remove_node). Parameter descriptions are present but often minimal. Output schemas are not documented. Error handling guidance is absent. The server follows a semantic graph model well but falls short of production-grade tool definition standards.

• Add descriptions to all 5 undescribed tools. For get_graph_size: 'Retrieve the total number of nodes in the graph. Call this first to decide whether to fetch the full graph (small) or summary (large >200 nodes) to avoid context bloat.' For get_neighborhood: 'Retrieve a subgraph containing a node and its neighbors up to a specified depth. Useful for exploring local context around a specific node without loading the entire graph.'
• Document output schemas for all tools. Example for add_node: 'Returns {node_index: usize, created_at: ISO8601 timestamp}'. For get_graph_full: 'Returns {dot_string: string, node_count: usize, edge_count: usize}'.
• Expand parameter descriptions to include constraints and examples. For node_index: 'The zero-based index of the node (0 to node_count-1). Returned by add_node.' For depth in get_neighborhood: 'Traversal depth (1-5 recommended; higher values may return large subgraphs).'
• Add error handling guidance to tool descriptions. Example: 'If node_index is out of bounds, returns error with available node count. If graph file is corrupted, returns error with recovery path (reinitialize from backup).'
• Implement a confirm_remove_node tool or add a dry_run parameter to remove_node. Return what would be deleted (node name, edge count) before actual deletion, allowing agents to verify intent.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry