28 KiB
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
Project Overview
n8n-mcp is a comprehensive documentation and knowledge server that provides AI assistants with complete access to n8n node information through the Model Context Protocol (MCP). It serves as a bridge between n8n's workflow automation platform and AI models, enabling them to understand and work with n8n nodes effectively.
Current Architecture:
src/
├── loaders/
│ └── node-loader.ts # NPM package loader for both packages
├── parsers/
│ ├── node-parser.ts # Enhanced parser with version support
│ └── property-extractor.ts # Dedicated property/operation extraction
├── mappers/
│ └── docs-mapper.ts # Documentation mapping with fixes
├── database/
│ ├── schema.sql # SQLite schema
│ ├── node-repository.ts # Data access layer
│ └── database-adapter.ts # Universal database adapter (NEW in v2.3)
├── services/
│ ├── property-filter.ts # Filters properties to essentials (NEW in v2.4)
│ ├── example-generator.ts # Generates working examples (NEW in v2.4)
│ ├── task-templates.ts # Pre-configured node settings (NEW in v2.4)
│ ├── config-validator.ts # Configuration validation (NEW in v2.4)
│ ├── enhanced-config-validator.ts # Operation-aware validation (NEW in v2.4.2)
│ ├── node-specific-validators.ts # Node-specific validation logic (NEW in v2.4.2)
│ ├── property-dependencies.ts # Dependency analysis (NEW in v2.4)
│ ├── expression-validator.ts # n8n expression syntax validation (NEW in v2.5.0)
│ └── workflow-validator.ts # Complete workflow validation (NEW in v2.5.0)
├── templates/
│ ├── template-fetcher.ts # Fetches templates from n8n.io API (NEW in v2.4.1)
│ ├── template-repository.ts # Template database operations (NEW in v2.4.1)
│ └── template-service.ts # Template business logic (NEW in v2.4.1)
├── scripts/
│ ├── rebuild.ts # Database rebuild with validation
│ ├── validate.ts # Node validation
│ ├── test-nodes.ts # Critical node tests
│ ├── test-essentials.ts # Test new essentials tools (NEW in v2.4)
│ ├── test-enhanced-validation.ts # Test enhanced validation (NEW in v2.4.2)
│ ├── test-workflow-validation.ts # Test workflow validation (NEW in v2.5.0)
│ ├── test-ai-workflow-validation.ts # Test AI workflow validation (NEW in v2.5.1)
│ ├── test-mcp-tools.ts # Test MCP tool enhancements (NEW in v2.5.1)
│ ├── test-n8n-validate-workflow.ts # Test n8n_validate_workflow tool (NEW in v2.6.3)
│ ├── test-typeversion-validation.ts # Test typeVersion validation (NEW in v2.6.1)
│ ├── test-workflow-diff.ts # Test workflow diff engine (NEW in v2.7.0)
│ ├── test-tools-documentation.ts # Test tools documentation (NEW in v2.7.3)
│ ├── fetch-templates.ts # Fetch workflow templates from n8n.io (NEW in v2.4.1)
│ └── test-templates.ts # Test template functionality (NEW in v2.4.1)
├── mcp/
│ ├── server.ts # MCP server with enhanced tools
│ ├── tools.ts # Tool definitions including new essentials
│ ├── tools-documentation.ts # Tool documentation system (NEW in v2.7.3)
│ └── index.ts # Main entry point with mode selection
├── utils/
│ ├── console-manager.ts # Console output isolation (NEW in v2.3.1)
│ └── logger.ts # Logging utility with HTTP awareness
├── http-server-single-session.ts # Single-session HTTP server (NEW in v2.3.1)
├── mcp-engine.ts # Clean API for service integration (NEW in v2.3.1)
└── index.ts # Library exports
Key Metrics:
- 525 nodes successfully loaded (100%) - Updated to n8n v1.97.1
- 520 nodes with properties (99%)
- 334 nodes with operations (63.6%)
- 457 nodes with documentation (87%)
- 263 AI-capable tools detected (major increase)
- All critical nodes pass validation
Key Commands
# Development
npm install # Install dependencies
npm run build # Build TypeScript (required before running)
npm run dev # Run in development mode with auto-reload
npm test # Run Jest tests
npm run typecheck # TypeScript type checking
npm run lint # Check TypeScript types (alias for typecheck)
# Core Commands:
npm run rebuild # Rebuild node database
npm run rebuild:optimized # Build database with embedded source code
npm run validate # Validate critical nodes
npm run test-nodes # Test critical node properties/operations
# Template Commands:
npm run fetch:templates # Fetch workflow templates from n8n.io (manual)
npm run fetch:templates:robust # Robust template fetching with retries
npm run test:templates # Test template functionality
# Test Commands:
npm run test:essentials # Test new essentials tools
npm run test:enhanced-validation # Test enhanced validation
npm run test:ai-workflow-validation # Test AI workflow validation
npm run test:mcp-tools # Test MCP tool enhancements
npm run test:single-session # Test single session HTTP
npm run test:template-validation # Test template validation
npm run test:n8n-manager # Test n8n management tools integration
npm run test:n8n-validate-workflow # Test n8n_validate_workflow tool
npm run test:typeversion-validation # Test typeVersion validation
npm run test:workflow-diff # Test workflow diff engine
npm run test:tools-documentation # Test MCP tools documentation system
# Workflow Validation Commands:
npm run test:workflow-validation # Test workflow validation features
# Dependency Update Commands:
npm run update:n8n:check # Check for n8n updates (dry run)
npm run update:n8n # Update n8n packages to latest versions
# HTTP Server Commands:
npm run start:http # Start server in HTTP mode
npm run start:http:fixed # Start with fixed HTTP implementation
npm run start:http:legacy # Start with legacy HTTP server
npm run http # Build and start HTTP server
npm run dev:http # HTTP server with auto-reload
# Legacy Commands (deprecated):
npm run db:rebuild # Old rebuild command
npm run db:init # Initialize empty database
npm run docs:rebuild # Rebuild documentation from TypeScript source
# Production
npm start # Run built application (stdio mode)
npm run start:http # Run in HTTP mode for remote access
# Docker Commands:
docker compose up -d # Start with Docker Compose
docker compose logs -f # View logs
docker compose down # Stop containers
docker compose down -v # Stop and remove volumes
./scripts/test-docker.sh # Test Docker deployment
Docker Deployment
The project includes ultra-optimized Docker support with NO n8n dependencies at runtime:
🚀 Key Optimization: Runtime-Only Dependencies
Important: Since the database is always pre-built before deployment, the Docker image contains NO n8n dependencies. This results in:
- 82% smaller images (~280MB vs ~1.5GB)
- 10x faster builds (~1-2 minutes vs ~12 minutes)
- No n8n version conflicts at runtime
- Minimal attack surface for security
Quick Start with Docker
# IMPORTANT: Rebuild database first (requires n8n locally)
npm run rebuild
# Create .env file with auth token
echo "AUTH_TOKEN=$(openssl rand -base64 32)" > .env
# Start the server
docker compose up -d
# Check health
curl http://localhost:3000/health
Docker Architecture
The Docker image contains ONLY these runtime dependencies:
@modelcontextprotocol/sdk- MCP protocol implementationbetter-sqlite3/sql.js- SQLite database accessexpress- HTTP server modedotenv- Environment configuration
Docker Features
- Ultra-optimized size (~280MB runtime-only)
- No n8n dependencies in production image
- Pre-built database required (nodes.db)
- BuildKit optimizations for fast builds
- Non-root user execution for security
- Health checks built into the image
Docker Images
ghcr.io/czlonkowski/n8n-mcp:latest- Runtime-only production image- Multi-architecture support (amd64, arm64)
- ~280MB compressed size (82% smaller!)
Docker Development
# Use BuildKit compose for development
COMPOSE_DOCKER_CLI_BUILD=1 docker-compose -f docker-compose.buildkit.yml up
# Build with optimizations
./scripts/build-optimized.sh
# Run tests
./scripts/test-docker.sh
For detailed Docker documentation, see DOCKER_README.md.
High-Level Architecture
The project implements MCP (Model Context Protocol) to expose n8n node documentation, source code, and examples to AI assistants. Key architectural components:
Core Services
- NodeDocumentationService (
src/services/node-documentation-service.ts): Main database service using SQLite with FTS5 for fast searching - MCP Server (
src/mcp/server.ts): Implements MCP protocol with tools for querying n8n nodes - Node Source Extractor (
src/utils/node-source-extractor.ts): Extracts node implementations from n8n packages - Enhanced Documentation Fetcher (
src/utils/enhanced-documentation-fetcher.ts): Fetches and parses official n8n documentation
MCP Tools Available
list_nodes- List all available n8n nodes with filteringget_node_info- Get comprehensive information about a specific node (now includes aiToolCapabilities)get_node_essentials- NEW Get only essential properties (10-20) with examples (95% smaller)get_node_as_tool_info- NEW v2.5.1 Get specific information about using ANY node as an AI toolsearch_nodes- Full-text search across all node documentationsearch_node_properties- NEW Search for specific properties within a nodeget_node_for_task- NEW Get pre-configured node settings for common taskslist_tasks- NEW List all available task templatesvalidate_node_operation- NEW v2.4.2 Verify node configuration with operation awareness and profilesvalidate_node_minimal- NEW v2.4.2 Quick validation for just required fieldsvalidate_workflow- NEW v2.5.0 Validate entire workflows before deployment (now validates ai_tool connections)validate_workflow_connections- NEW v2.5.0 Check workflow structure and connectionsvalidate_workflow_expressions- NEW v2.5.0 Validate all n8n expressions in a workflowget_property_dependencies- NEW Analyze property dependencies and visibility conditionslist_ai_tools- List all AI-capable nodes (now includes usage guidance)get_node_documentation- Get parsed documentation from n8n-docsget_database_statistics- Get database usage statistics and metricslist_node_templates- NEW Find workflow templates using specific nodesget_template- NEW Get complete workflow JSON for importsearch_templates- NEW Search templates by keywordsget_templates_for_task- NEW Get curated templates for common taskstools_documentation- NEW v2.7.3 Get comprehensive documentation for MCP tools
n8n Management Tools (NEW v2.6.0 - Requires API Configuration)
These tools are only available when N8N_API_URL and N8N_API_KEY are configured:
Workflow Management
n8n_create_workflow- Create new workflows with nodes and connectionsn8n_get_workflow- Get complete workflow by IDn8n_get_workflow_details- Get workflow with execution statisticsn8n_get_workflow_structure- Get simplified workflow structuren8n_get_workflow_minimal- Get minimal workflow infon8n_update_full_workflow- Update existing workflows (complete replacement)n8n_update_partial_workflow- NEW v2.7.0 Update workflows using diff operationsn8n_delete_workflow- Delete workflows permanentlyn8n_list_workflows- List workflows with filteringn8n_validate_workflow- NEW v2.6.3 Validate workflow from n8n instance by ID
Execution Management
n8n_trigger_webhook_workflow- Trigger workflows via webhook URLn8n_get_execution- Get execution details by IDn8n_list_executions- List executions with status filteringn8n_delete_execution- Delete execution records
System Tools
n8n_health_check- Check n8n API connectivity and featuresn8n_list_available_tools- List all available management tools
Database Structure
Uses SQLite with enhanced schema:
- nodes table: Core node information with FTS5 indexing
- node_documentation: Parsed markdown documentation
- node_examples: Generated workflow examples
- node_source_code: Complete TypeScript/JavaScript implementations
Important Development Notes
Initial Setup Requirements
- Clone n8n-docs:
git clone https://github.com/n8n-io/n8n-docs.git ../n8n-docs - Install Dependencies:
npm install - Build:
npm run build - Rebuild Database:
npm run rebuild - Validate:
npm run test-nodes
Key Technical Decisions (v2.3)
-
Database Adapter Implementation:
- Created
DatabaseAdapterinterface to abstract database operations - Implemented
BetterSQLiteAdapterandSQLJSAdapterclasses - Used factory pattern in
createDatabaseAdapter()for automatic selection - Added persistence layer for sql.js with debounced saves (100ms)
- Created
-
Compatibility Strategy:
- Primary: Try better-sqlite3 first for performance
- Fallback: Catch native module errors and switch to sql.js
- Detection: Check for NODE_MODULE_VERSION errors specifically
- Logging: Clear messages about which adapter is active
-
Performance Considerations:
- better-sqlite3: ~10-50x faster for most operations
- sql.js: ~2-5x slower but acceptable for this use case
- Auto-save: 100ms debounce prevents excessive disk writes with sql.js
- Memory: sql.js uses more memory but manageable for our dataset size
Node.js Version Compatibility
The project now features automatic database adapter fallback for universal Node.js compatibility:
- Primary adapter: Uses
better-sqlite3for optimal performance when available - Fallback adapter: Automatically switches to
sql.js(pure JavaScript) if:- Native modules fail to load
- Node.js version mismatch detected
- Running in Claude Desktop or other restricted environments
This means the project works with ANY Node.js version without manual intervention. The adapter selection is automatic and transparent.
Implementation Status
- ✅ Property/operation extraction for 98.7% of nodes
- ✅ Support for both n8n-nodes-base and @n8n/n8n-nodes-langchain
- ✅ AI tool detection (35 tools with usableAsTool property)
- ✅ Versioned node support (HTTPRequest, Code, etc.)
- ✅ Documentation coverage for 88.6% of nodes
- ⏳ Version history tracking (deferred - only current version)
- ⏳ Workflow examples (deferred - using documentation)
Testing Workflow
npm run build # Always build first
npm test # Run all tests
npm run typecheck # Verify TypeScript types
Docker Development
# Local development with stdio
docker-compose -f docker-compose.local.yml up
# HTTP server mode
docker-compose -f docker-compose.http.yml up
Authentication (HTTP mode)
When running in HTTP mode, use Bearer token authentication:
Authorization: Bearer your-auth-token
Architecture Patterns
Service Layer Pattern
All major functionality is implemented as services in src/services/. When adding new features:
- Create a service class with clear responsibilities
- Use dependency injection where appropriate
- Implement proper error handling with custom error types
- Add comprehensive logging using the logger utility
MCP Tool Implementation
When adding new MCP tools:
- Define the tool in
src/mcp/tools.ts - Implement handler in
src/mcp/server.ts - Add proper input validation
- Return structured responses matching MCP expectations
Database Access Pattern
- Use prepared statements for all queries
- Implement proper transaction handling
- Use FTS5 for text searching
- Cache frequently accessed data in memory
Database Adapter Pattern (NEW in v2.3)
The project uses a database adapter pattern for universal compatibility:
- Primary adapter:
better-sqlite3- Native SQLite bindings for optimal performance - Fallback adapter:
sql.js- Pure JavaScript implementation for compatibility - Automatic selection: The system detects and handles version mismatches automatically
- Unified interface: Both adapters implement the same
DatabaseAdapterinterface - Transparent operation: Application code doesn't need to know which adapter is active
Environment Configuration
Required environment variables (see .env.example):
# Server Configuration
NODE_ENV=development
PORT=3000
AUTH_TOKEN=your-secure-token
# Trust proxy for correct IP logging (optional)
# Set to 1 when behind a reverse proxy (Nginx, etc.)
TRUST_PROXY=0
# MCP Configuration
MCP_SERVER_NAME=n8n-documentation-mcp
MCP_SERVER_VERSION=1.0.0
# Logging
LOG_LEVEL=info
License
This project is licensed under the MIT License. Created by Romuald Czlonkowski @ www.aiadvisors.pl/en.
- ✅ Free for any use (personal, commercial, etc.)
- ✅ Modifications and distribution allowed
- ✅ Can be included in commercial products
- ✅ Can be hosted as a service
Attribution is appreciated but not required. See LICENSE and ATTRIBUTION.md for details.
HTTP Remote Deployment (v2.3.0)
✅ HTTP Server Implementation Complete
The project now includes a simplified HTTP server mode for remote deployments:
- Single-user design: Stateless architecture for private deployments
- Simple token auth: Bearer token authentication
- MCP-compatible: Works with mcp-remote adapter for Claude Desktop
- Easy deployment: Minimal configuration required
Quick Start
# Server setup
export MCP_MODE=http
export AUTH_TOKEN=$(openssl rand -base64 32)
npm run start:http
# Client setup (Claude Desktop config)
{
"mcpServers": {
"n8n-remote": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/mcp-remote@latest",
"connect",
"https://your-server.com/mcp"
],
"env": {
"MCP_AUTH_TOKEN": "your-auth-token"
}
}
}
}
Available Scripts
npm run start:http- Start in HTTP modenpm run http- Build and start HTTP servernpm run dev:http- Development mode with auto-reload./scripts/deploy-http.sh- Deployment helper script
For detailed deployment instructions, see HTTP Deployment Guide.
Recent Problem Solutions
MCP HTTP Server Errors (Solved in v2.3.2)
Problem: Two critical errors prevented the HTTP server from working:
- "stream is not readable" - Express.json() middleware consumed the request stream
- "Server not initialized" - StreamableHTTPServerTransport initialization issues
Solution: Two-phase fix:
- Removed body parsing middleware to preserve raw stream
- Created direct JSON-RPC implementation bypassing StreamableHTTPServerTransport
Technical Details:
src/http-server-single-session.ts- Single-session implementation (partial fix)src/http-server.ts- Direct JSON-RPC implementation (complete fix)src/utils/console-manager.ts- Console output isolation- Use
USE_FIXED_HTTP=trueto enable the fixed implementation
SQLite Version Mismatch (Solved in v2.3)
Problem: Claude Desktop bundles Node.js v16.19.1, causing NODE_MODULE_VERSION errors with better-sqlite3 compiled for different versions.
Solution: Implemented dual-adapter system:
- Database adapter abstraction layer
- Automatic fallback from better-sqlite3 to sql.js
- Transparent operation regardless of Node.js version
- No manual configuration required
Technical Details:
src/database/database-adapter.ts- Adapter interface and implementationscreateDatabaseAdapter()- Factory function with automatic selection- Modified all database operations to use adapter interface
- Added sql.js with persistence support
Property Extraction Issues (Solved in v2.2)
Problem: Many nodes had empty properties/operations arrays.
Solution: Created dedicated PropertyExtractor class that handles:
- Instance-level property extraction
- Versioned node support
- Both programmatic and declarative styles
- Complex nested property structures
Dependency Update Issues (Solved in v2.3.3)
Problem: n8n packages have interdependent version requirements. Updating them independently causes version mismatches.
Solution: Implemented smart dependency update system:
- Check n8n's required dependency versions
- Update all packages to match n8n's requirements
- Validate database after updates
- Fix node type references in validation script
Technical Details:
scripts/update-n8n-deps.js- Smart dependency updater.github/workflows/update-n8n-deps.yml- GitHub Actions automationrenovate.json- Alternative Renovate configuration- Fixed validation to use 'nodes-base.httpRequest' format instead of 'httpRequest'
AI-Optimized Tools (NEW in v2.4.0)
Problem: get_node_info returns 100KB+ of JSON with 200+ properties, making it nearly impossible for AI agents to efficiently configure nodes.
Solution: Created new tools that provide progressive disclosure of information:
get_node_essentials- Returns only the 10-20 most important propertiessearch_node_properties- Find specific properties without downloading everything
Results:
- 95% reduction in response size (100KB → 5KB)
- Only essential and commonly-used properties returned
- Includes working examples for immediate use
- AI agents can now configure nodes in seconds instead of minutes
Technical Implementation:
src/services/property-filter.ts- Curated essential properties for 20+ nodessrc/services/example-generator.ts- Working examples for common use cases- Smart property search with relevance scoring
- Automatic fallback for unconfigured nodes
Usage Recommendation:
# OLD approach (avoid):
get_node_info("nodes-base.httpRequest") # 100KB+ response
# NEW approach (prefer):
get_node_essentials("nodes-base.httpRequest") # 5KB response with examples
Task-Based Configuration (NEW in v2.4.0)
Problem: AI agents need to know exactly how to configure nodes for common tasks like "send email", "fetch API", or "update database".
Solution: Created task template system:
- Pre-configured node settings for common tasks
- Working examples with proper credentials structure
- Task discovery via
list_taskstool
Results:
- Instant node configuration for common tasks
- No guessing about property values
- Production-ready configurations
- Covers 30+ common automation tasks
Workflow Template Support (NEW in v2.4.1)
Problem: AI agents needed complete workflow examples to understand how nodes work together.
Solution: Integrated n8n.io workflow templates:
- 10,000+ templates available via MCP tools
- Search by keywords or node usage
- Get complete workflow JSON for import
- Task-based template suggestions
Technical Details:
- Templates fetched from official n8n.io API
- Stored in SQLite with FTS5 search
- Includes metadata: categories, node counts, user ratings
- Smart caching to prevent API overload
Enhanced Validation with Profiles (NEW in v2.4.2)
Problem: Different validation needs - quick checks during editing vs thorough validation before deployment.
Solution: Validation profiles with operation awareness:
- strict - Full validation (deployment)
- standard - Common issues only (default)
- minimal - Just required fields
- quick - Fast essential checks
Results:
- 90% faster validation for editing workflows
- Operation-specific validation rules
- Better error messages with fix suggestions
- Node-specific validators for complex nodes
Complete Workflow Validation (NEW in v2.5.0)
Problem: Node validation wasn't enough - needed to validate entire workflows including connections, expressions, and dependencies.
Solution: Three-layer workflow validation:
- Structure validation - Nodes, connections, dependencies
- Configuration validation - All node configs with operation awareness
- Expression validation - n8n expression syntax checking
Results:
- Catch workflow errors before deployment
- Validate complex multi-node workflows
- Check all n8n expressions for syntax errors
- Ensure proper node connections and data flow
AI Tool Support Enhancement (NEW in v2.5.1)
Problem: AI agents needed better guidance on using n8n nodes as AI tools and understanding tool connections.
Solution: Enhanced AI tool support:
- New
get_node_as_tool_info- Explains how ANY node can be used as an AI tool - Enhanced workflow validation for ai_tool node connections
- Better documentation for AI tool usage patterns
- Validation ensures proper tool node connections
Results:
- AI agents can now properly configure AI tool workflows
- Clear guidance on credential requirements for tools
- Validation catches common AI workflow mistakes
- Supports both native AI nodes and regular nodes as tools
n8n Management Integration (NEW in v2.6.0)
Problem: AI agents could discover and validate workflows but couldn't deploy or execute them.
Solution: Integrated n8n-manager-for-ai-agents functionality:
- 14 new management tools when API configured
- Complete workflow lifecycle support
- Smart error handling for API limitations
- Optional feature - only loads when configured
Results:
- Full workflow automation: discover → build → validate → deploy → execute
- Webhook-based workflow triggering
- Execution monitoring and management
- Backwards compatible - doesn't affect existing functionality
Workflow Diff Engine (NEW in v2.7.0)
Problem: Updating workflows required sending the entire JSON (often 50KB+), wasting tokens and making it hard to see what changed.
Solution: Diff-based workflow updates:
- 13 targeted operations - Add, remove, update, move nodes/connections
- 80-90% token savings - Only send the changes
- Transactional updates - All changes validated before applying
- Order independence - Add connections before nodes exist
Results:
- Update a single node property without sending entire workflow
- Clear audit trail of what changed
- Safer updates with validation
- Works with any workflow size
Known Issues
Claude Desktop - Duplicate Container Bug
When adding n8n-mcp to Claude Desktop, you might see "Container with name '/n8n-mcp-container' already exists" error. This is a Claude Desktop bug where it doesn't properly clean up containers between sessions.
Workaround: Add this to your Claude Desktop config to use a unique container name each time:
{
"command": "docker",
"args": [
"run",
"--rm",
"--name", "n8n-mcp-{{timestamp}}",
"-e", "AUTH_TOKEN=your-token",
"ghcr.io/czlonkowski/n8n-mcp:latest"
]
}
Note: {{timestamp}} is not actually supported by Claude Desktop. The real workaround is to manually remove the container when this happens:
docker rm n8n-mcp-container
See Issue #13 for more details.
npm Publishing
To publish a new version to npm:
# 1. Update version in package.json
npm version patch # or minor/major
# 2. Prepare the publish directory
npm run prepare:publish
# 3. Publish to npm (requires OTP)
cd npm-publish-temp
npm publish --otp=YOUR_OTP_CODE
# 4. Clean up
cd ..
rm -rf npm-publish-temp
The published package can then be used with npx:
npx n8n-mcp
Future Enhancements
Planned Features
- Historical version tracking for nodes
- Workflow template generation from examples
- Performance metrics and optimization suggestions
- Integration with n8n Cloud API for live data
- WebSocket support for real-time updates
Contributing
Contributions are welcome! Please:
- Follow the existing code patterns
- Add tests for new functionality
- Update documentation as needed
- Run all tests before submitting PRs
For questions or support, please open an issue on GitHub.