Skip to content

07_mcp_tools.md

Latest commit

 

History

History
505 lines (411 loc) · 13.9 KB

07_mcp_tools.md

File metadata and controls

505 lines (411 loc) · 13.9 KB
layout default
title Chapter 7: MCP Tools Architecture
parent n8n-MCP Tutorial
nav_order 7

Chapter 7: MCP Tools Architecture

Welcome to the tools layer—the heart of n8n-MCP's functionality! In Chapter 6, we explored how InstanceContext enables flexible multi-tenant deployments. Now let's dive into the MCP tools—the actual capabilities that AI assistants use to interact with n8n.

Think of MCP tools as a comprehensive toolkit that transforms AI assistants from helpful generalists into n8n workflow automation experts. Each tool is carefully designed to provide specific, actionable capabilities.

What Are MCP Tools?

MCP tools are the interface between AI assistants and external capabilities. They follow a standardized JSON-RPC format:

// Tool Definition
interface Tool {
  name: string;
  description: string;
  inputSchema: JSONSchema;
}

// Tool Call
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "tool_name",
    "arguments": {
      "param1": "value1",
      "param2": "value2"
    }
  }
}

// Tool Response
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "Tool execution result"
      }
    ]
  }
}

n8n-MCP Tool Categories

n8n-MCP organizes its 15+ tools into logical categories based on functionality:

flowchart TD
    A[MCP Tools] --> B[System Tools]
    A --> C[Discovery Tools]
    A --> D[Configuration Tools]
    A --> E[Validation Tools]
    A --> F[Template Tools]
    A --> G[Workflow Management Tools]

    B --> B1[tools_documentation]
    B --> B2[n8n_health_check]

    C --> C1[search_nodes]

    D --> D1[get_node]

    E --> E1[validate_node]
    E --> E2[validate_workflow]

    F --> F1[get_template]
    F --> F2[search_templates]

    G --> G1[n8n_create_workflow]
    G --> G2[n8n_get_workflow]
    G --> G3[n8n_update_full_workflow]
    G --> G4[n8n_update_partial_workflow]
    G --> G5[n8n_delete_workflow]
    G --> G6[n8n_list_workflows]
    G --> G7[n8n_validate_workflow]
    G --> G8[n8n_autofix_workflow]
    G --> G9[n8n_test_workflow]
    G --> G10[n8n_executions]
    G --> G11[n8n_workflow_versions]
    G --> G12[n8n_deploy_template]
Loading

Tool Design Principles

Each tool follows consistent design principles:

1. Single Responsibility

Every tool does one thing well. For example:

  • search_nodes: Only searches, doesn't retrieve details
  • get_node: Only retrieves details, doesn't search

2. Progressive Information Disclosure

Tools provide information at the right level of detail:

  • Search tools: Provide overviews for browsing
  • Get tools: Provide complete details for specific items
  • Management tools: Provide operational capabilities

3. Consistent Parameter Patterns

// All search tools follow similar patterns
interface SearchParams {
  query?: string;        // Text search
  limit?: number;        // Result limit (default: 10)
  offset?: number;       // Pagination offset (default: 0)
  filters?: Record<string, any>; // Category-specific filters
}

4. Structured Responses

All tools return consistent response formats with metadata:

interface ToolResponse {
  success: boolean;
  data: any;            // Tool-specific result
  metadata: {
    timestamp: string;
    executionTime: number;
    version: string;
    instanceId?: string;
  };
  errors?: string[];    // Only present if success = false
}

Tool Categories Deep Dive

System Tools - Housekeeping and Diagnostics

Purpose: Provide essential system information and diagnostics

tools_documentation

  • What it does: Lists all available MCP tools with descriptions
  • Use case: Help AI assistants discover what capabilities are available
  • Parameters: None
  • Response: Complete tool catalog with schemas

n8n_health_check

  • What it does: Checks n8n instance connectivity and health
  • Use case: Verify n8n connection before performing operations
  • Parameters: None
  • Response: Health status, version info, connectivity status

Discovery Tools - Finding n8n Components

Purpose: Help AI assistants find relevant n8n nodes and components

search_nodes

  • What it does: Search through 1,000+ n8n nodes using text and filters
  • Use case: "Find email-related nodes" or "Show me AI tools"
  • Parameters:
    • query: Text search across names, descriptions, categories
    • category: Filter by node category (Communication, Data, etc.)
    • is_ai_tool: Filter for AI-capable nodes
    • limit/offset: Pagination
  • Response: List of matching nodes with basic info

Configuration Tools - Understanding Node Details

Purpose: Provide detailed information about specific n8n components

get_node

  • What it does: Retrieve complete details for a specific node
  • Use case: "Tell me everything about the Gmail node"
  • Parameters:
    • node_type: Exact node type (e.g., "n8n-nodes-base.gmail")
  • Response: Complete node schema, properties, operations, documentation

Validation Tools - Ensuring Correctness

Purpose: Help AI assistants validate their work and catch errors

validate_node

  • What it does: Validate a node's configuration against its schema
  • Use case: "Check if my Gmail configuration is correct"
  • Parameters:
    • node_type: The node type to validate against
    • configuration: The node configuration object
  • Response: Validation results with specific errors and suggestions

validate_workflow

  • What it does: Validate complete workflow structure and connections
  • Use case: "Check if this workflow will work"
  • Parameters:
    • workflow: Complete workflow object with nodes and connections
  • Response: Comprehensive validation report

Template Tools - Learning from Examples

Purpose: Provide access to workflow templates and examples

get_template

  • What it does: Retrieve a specific workflow template
  • Use case: "Show me the email marketing template"
  • Parameters:
    • template_id: Unique template identifier
  • Response: Complete template with workflow data and metadata

search_templates

  • What it does: Search through 2,700+ workflow templates
  • Use case: "Find templates for social media automation"
  • Parameters:
    • query: Text search across template names and descriptions
    • category: Filter by template category
    • limit/offset: Pagination
  • Response: List of matching templates

Workflow Management Tools - Direct n8n Operations

Purpose: Enable AI assistants to create, modify, and manage n8n workflows

Core CRUD Operations

  • n8n_create_workflow: Create new workflows from scratch
  • n8n_get_workflow: Retrieve existing workflow details
  • n8n_update_full_workflow: Replace entire workflows
  • n8n_update_partial_workflow: Modify specific parts of workflows
  • n8n_delete_workflow: Remove workflows
  • n8n_list_workflows: Browse available workflows

Advanced Operations

  • n8n_validate_workflow: Check workflow validity
  • n8n_autofix_workflow: Automatically fix common workflow issues
  • n8n_test_workflow: Execute workflows for testing
  • n8n_executions: View workflow execution history
  • n8n_workflow_versions: Access workflow version history
  • n8n_deploy_template: Deploy templates as active workflows

Tool Implementation Architecture

Base Tool Structure

All tools extend a common base class for consistency:

abstract class BaseTool {
  abstract readonly name: string;
  abstract readonly description: string;
  abstract readonly inputSchema: JSONSchema;

  abstract execute(args: any, context: ToolContext): Promise<ToolResult>;

  protected validateArgs(args: any): ValidationResult {
    // Common validation logic
  }

  protected createSuccessResponse(data: any): ToolResult {
    return {
      success: true,
      data,
      metadata: {
        timestamp: new Date().toISOString(),
        executionTime: Date.now() - this.startTime,
        version: this.getVersion()
      }
    };
  }
}

Tool Context

Every tool execution receives contextual information:

interface ToolContext {
  instanceContext: InstanceContext;    // n8n connection details
  sessionId: string;                   // Current session identifier
  userId?: string;                     // User identifier (if available)
  requestId: string;                   // Unique request identifier
  startTime: number;                   // Execution start timestamp
}

Error Handling

Tools implement consistent error handling:

class ToolError extends Error {
  constructor(
    message: string,
    public readonly code: ToolErrorCode,
    public readonly details?: any
  ) {
    super(message);
  }
}

enum ToolErrorCode {
  VALIDATION_ERROR = 'VALIDATION_ERROR',
  N8N_API_ERROR = 'N8N_API_ERROR',
  NOT_FOUND = 'NOT_FOUND',
  PERMISSION_DENIED = 'PERMISSION_DENIED',
  RATE_LIMITED = 'RATE_LIMITED'
}

Tool Registration and Discovery

Tools are registered in a centralized registry:

class ToolRegistry {
  private tools = new Map<string, BaseTool>();

  register(tool: BaseTool): void {
    this.tools.set(tool.name, tool);
  }

  getTool(name: string): BaseTool | undefined {
    return this.tools.get(name);
  }

  getAllTools(): BaseTool[] {
    return Array.from(this.tools.values());
  }

  // Generate MCP tool list for protocol handshake
  getMCPToolList(): Tool[] {
    return this.getAllTools().map(tool => ({
      name: tool.name,
      description: tool.description,
      inputSchema: tool.inputSchema
    }));
  }
}

Performance Optimizations

Caching Strategies

class ToolCache {
  // Cache node data (changes infrequently)
  @Cache({ ttl: 3600000 }) // 1 hour
  async getNodeData(nodeType: string): Promise<NodeData> {
    return this.storage.getNode(nodeType);
  }

  // Cache template searches (changes moderately)
  @Cache({ ttl: 300000 }) // 5 minutes
  async searchTemplates(query: string): Promise<Template[]> {
    return this.storage.searchTemplates(query);
  }

  // Don't cache workflow operations (changes frequently)
  async getWorkflow(id: string): Promise<Workflow> {
    return this.n8nApi.getWorkflow(id);
  }
}

Batch Operations

class BatchToolProcessor {
  async executeBatch(operations: ToolOperation[]): Promise<ToolResult[]> {
    // Group operations by type for efficiency
    const grouped = this.groupByTool(operations);

    // Execute in parallel where possible
    const results = await Promise.allSettled(
      grouped.map(group => this.executeToolBatch(group))
    );

    return this.flattenResults(results);
  }
}

Security and Access Control

Tool-Level Permissions

interface ToolPermissions {
  read: boolean;      // Can use read-only tools
  write: boolean;     // Can use workflow modification tools
  execute: boolean;   // Can execute workflows
  admin: boolean;     // Can use administrative tools
}

class PermissionChecker {
  checkToolAccess(toolName: string, context: ToolContext): boolean {
    const permissions = this.getPermissions(context.userId);

    switch (toolName) {
      case 'n8n_create_workflow':
      case 'n8n_update_workflow':
        return permissions.write;

      case 'n8n_test_workflow':
        return permissions.execute;

      default:
        return permissions.read;
    }
  }
}

Rate Limiting

class ToolRateLimiter {
  private limits = new Map<string, RateLimit>();

  checkLimit(toolName: string, userId: string): boolean {
    const key = `${toolName}:${userId}`;
    const limit = this.limits.get(key) || this.getDefaultLimit(toolName);

    return limit.allow();
  }

  private getDefaultLimit(toolName: string): RateLimit {
    // Different limits for different tool types
    if (toolName.startsWith('n8n_')) {
      return new RateLimit(10, 'minute');  // 10 workflow ops/minute
    } else {
      return new RateLimit(60, 'minute');  // 60 search ops/minute
    }
  }
}

Tool Development Workflow

1. Define Tool Interface

interface SearchNodesArgs {
  query?: string;
  category?: string;
  is_ai_tool?: boolean;
  limit?: number;
  offset?: number;
}

2. Implement Tool Class

class SearchNodesTool extends BaseTool {
  readonly name = 'search_nodes';
  readonly description = 'Search through n8n nodes using text and filters';

  readonly inputSchema = {
    type: 'object',
    properties: {
      query: { type: 'string' },
      category: { type: 'string' },
      is_ai_tool: { type: 'boolean' },
      limit: { type: 'number', default: 10 },
      offset: { type: 'number', default: 0 }
    }
  };

  async execute(args: SearchNodesArgs, context: ToolContext): Promise<ToolResult> {
    // Implementation here
  }
}

3. Register and Test

// Register tool
registry.register(new SearchNodesTool());

// Test tool
const result = await registry.getTool('search_nodes')?.execute({
  query: 'gmail',
  limit: 5
}, context);

Tool Analytics and Monitoring

Usage Tracking

class ToolAnalytics {
  recordToolUsage(toolName: string, context: ToolContext, success: boolean, duration: number) {
    // Store in database for analysis
    this.db.insert('tool_usage', {
      tool_name: toolName,
      user_id: context.userId,
      instance_id: context.instanceContext.instanceId,
      success,
      duration,
      timestamp: new Date()
    });
  }

  getUsageStats(toolName?: string): UsageStats {
    // Generate usage reports
  }
}

Congratulations! You now understand the comprehensive MCP tools architecture that powers n8n-MCP. These tools transform AI assistants into n8n workflow automation experts.

In the next chapters, we'll explore specific tool categories in detail: discovery tools for finding nodes and workflow management tools for direct n8n operations.