MCP server that lets agents talk to Workflowy
Strong tool naming (verb_noun pattern), comprehensive parameter descriptions with practical guidance (e.g., 'Use maxDepth=10 for entire subtree'), and clear risk classifications. All 8 tools have descriptions and input schemas with types. However, output schemas are not documented, LLMs cannot predict response structure. Batch operations are well-designed for composition. Error handling lacks recovery guidance (no 'try search_nodes first' hints). Parameter descriptions exceed baseline length (194 chars avg) but are actionable. Missing: explicit output field documentation, error classification, and confirmation patterns for destructive ops.
Create multiple sibling nodes atomically (all succeed or none). PREFERRED for 2+ nodes vs multiple create_node calls.
Delete multiple nodes atomically. PREFERRED for 2+ nodes vs multiple delete_node calls.
Update multiple nodes atomically. PREFERRED for 2+ nodes vs multiple update_node calls.
Create a single node under a parent. For 2+ nodes, use batch_create_nodes (more efficient).
Delete a single node. WARNING: Deletes all descendants (cascade). For 2+ nodes, use batch_delete_nodes.
List child nodes from a parent (or root if no parentId). EFFICIENCY: Use maxDepth=10 to get entire subtree in one call. Default maxDepth=0 returns only direct children.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Forces agents to guess what list_nodes returns.
Destructive operations (delete_node, batch_delete_nodes) lack confirmation/dry-run pattern. Agents can cascade-delete entire subtrees without safeguard.
Error responses lack recovery guidance. No hints like 'If node not found, try search_nodes()' to guide agent retry logic.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
Search nodes by text across the entire tree. Use maxDepth=2 to include parent/child context with matches.
Update a single node. For 2+ nodes, use batch_update_nodes.
Batch operations (batch_create_nodes, batch_update_nodes, batch_delete_nodes) lack per-item success/failure reporting. One failed item fails entire batch.
Parameter 'preview' truncates content but no guidance on when to use it. Agents may not understand token-saving implications.