Loading ToolBench
Advanced MCP server for n8n workflow creation, optimization, and management
The server provides 22 tools with reasonable descriptions and schemas, but exhibits several critical gaps in naming clarity, parameter documentation, and output schema definition. Tool names generally start with action verbs (add_, get_, sync_, find_, etc.) which is positive, but lack consistency in distinguishing similar operations (e.g., get_template_by_id vs extract_template_intent both retrieve templates but with unclear differentiation). Descriptions range from 100-350 characters, generally adequate, but parameter documentation is incomplete, many tools lack descriptions for optional parameters (e.g., 'sync_templates' has no description for its optional 'force' parameter). Output schemas are not documented in the provided source, making it unclear what fields agents should expect from responses. Error handling guidance is absent. Schema validation is present for inputs but incomplete, several tools have parameters without type specifications or constraints. The 'intent' tool family (add_node_intent through remove_node_intent) shows good domain-specific naming and purpose clarity. Template tools suffer from overlapping responsibility (sync_templates, get_template_stats, get_popular_templates all manage or introspect the same template library). No evidence of secrets handling, audit trails, or permission gates. Overall pattern adherence is moderate, the server implements basic tool structure but lacks depth in parameter guidance, output clarity, and error recovery paths expected of production tools.
Adapt a template to specific requirements
📝 Add 'why' metadata to a workflow node for AI context continuity. Helps LLMs remember the reasoning, assumptions, and risks across iterations. This is crucial for maintaining design decisions over time.
Analyze impact of changes to a workflow
📊 Analyze how well a workflow is documented with intent metadata. Shows coverage percentage, identifies nodes without intent, and provides recommendations for improving documentation.
Check workflow compatibility
Clear template cache
Compare two workflows and show differences
Parameter descriptions missing or trivial for 12+ tools. 'Sync templates from configured sources' for sync_templates, 'Get template library statistics' for get_template_stats, these descriptions do not explain WHEN to use each tool, what it returns, or how it differs from similar tools (get_popular_templates, get_recent_templates both browse templates but purpose is unclear).
No output schema documentation. The source code shows input schemas clearly (e.g., add_node_intent has well-defined inputSchema), but there is NO specification of what fields, types, or structure agents should expect in responses. LLMs cannot plan downstream calls (e.g., 'pass the template_id to adapt_template') without knowing what fields are returned.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | 2026-07-28+ | v2 |
| 2026-03-09 | D | 54 | - | v1 |
Analyze workflows to discover node types
Extract intent from a template
Find templates matching a natural language intent
Get schema information for a discovered node type
Get most popular templates
Get recently added templates
Get detailed template information
Get provenance information for a template
Get requirements for a template
Get template library statistics
📋 Get all intent metadata from a workflow. Shows the 'why' behind each node - perfect for understanding existing workflows or resuming work after a break.
🗑️ Remove intent metadata from a node. Use when intent is no longer relevant or needs to be completely rewritten.
💡 Get AI-generated intent template for a specific node. Provides a starting point for documenting the 'why' based on node type and context. Saves time and ensures consistent documentation.
Sync templates from configured sources
✏️ Update existing intent metadata for a node. Use this to refine the documentation as understanding evolves or circumstances change.
Optional parameters lack descriptions. sync_templates has a 'force' boolean (line: 'force: {type: boolean, description: Force sync...}') but intent is unclear, when should an agent set it? What is 'recently synced'? The parameter description does not explain the consequence of setting force=true.
Overlapping tool responsibilities with no disambiguation. get_popular_templates, get_recent_templates, and find_templates_by_intent all retrieve templates from the library. The descriptions ('Get most popular templates' vs 'Get recently added templates' vs 'Find templates matching...') are too similar, an LLM cannot distinguish when to call each one. Recommend consolidating into search_templates with a 'sort_by' enum (popular|recent|relevance) or separate ONLY if the underlying APIs are truly distinct.
No error handling guidance. No tool description explains what to do if a workflow_id is not found, a template_id is invalid, or a sync fails. Responses lack actionable error recovery hints (e.g., 'Workflow not found. Available workflows: [list]. Did you mean...?'). This forces LLMs to retry blindly or abandon.
Destructive operations (remove_node_intent, adapt_template, sync_templates with force=true) lack confirmation or dry-run capability. An agent could irreversibly delete node metadata or overwrite templates without explicit user approval. Missing confirmation-request pattern.
Schema constraints incomplete. Several parameters lack enum, range, or pattern specifications. For example, get_workflow_intents has 'format' with enum ['report', 'json'], which is good, but discover_nodes has an empty properties object, no parameters at all are described, making it a black box. adapt_template's 'customizations' is declared as generic 'object' with no schema for nested fields, LLMs cannot reason about what keys to include.
No pagination or result limits documented. Tools like get_popular_templates, find_templates_by_intent, and discover_nodes do not specify max result counts. If a template library has 10,000 templates, an agent could retrieve all of them in a single call, blowing the context window. Descriptions do not mention limits, pagination, or top_k constraints.
No audit trail, permission gates, or secret injection patterns. The source code provides no evidence of logging who called what tool, checking permissions, or injecting secrets server-side. Destructive and sensitive tools (remove_node_intent, sync_templates, adapt_template) execute without permission verification or audit trails, violating audit-trail and permission-gate patterns.
Tool naming ambiguity in template family. 'get_template_by_id' suggests retrieval by ID, but 'extract_template_intent' also retrieves template data (extracting intent from it). These names do not make clear that they return different fields or serve different purposes. Consider 'get_template_details' vs 'get_template_intent' or add disambiguating descriptions.