## Problem
AI assistants were consistently connecting SplitInBatches node outputs backwards because:
- Output index 0 = "done" (runs after loop completes)
- Output index 1 = "loop" (processes items inside loop)
This counterintuitive ordering caused incorrect workflow connections.
## Solution
Enhanced the n8n-mcp system to expose and clarify output information:
### Database & Schema
- Added `outputs` and `output_names` columns to nodes table
- Updated NodeRepository to store/retrieve output information
### Node Parsing
- Enhanced NodeParser to extract outputs and outputNames from nodes
- Properly handles versioned nodes like SplitInBatchesV3
### MCP Server
- Modified getNodeInfo to return detailed output descriptions
- Added connection guidance for each output
- Special handling for loop nodes (SplitInBatches, IF, Switch)
### Documentation
- Enhanced DocsMapper to inject critical output guidance
- Added warnings about counterintuitive output ordering
- Provides correct connection patterns for loop nodes
### Workflow Validation
- Added validateSplitInBatchesConnection method
- Detects reversed connections and provides specific errors
- Added checkForLoopBack with depth limit to prevent stack overflow
- Smart heuristics to identify likely connection mistakes
## Testing
- Created comprehensive test suite (81 tests)
- Unit tests for all modified components
- Edge case handling for malformed data
- Performance testing with large workflows
## Impact
AI assistants will now:
- See explicit output indices and names (e.g., "Output 0: done")
- Receive clear connection guidance
- Get validation errors when connections are reversed
- Have enhanced documentation explaining the correct pattern
Fixes#97🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Added cleanupTimer property to track setInterval timer
- Implemented destroy() method to clear timer and prevent memory leak
- Updated MCP server shutdown to call cache.destroy()
- Enhanced HTTP server error handling with transport.onerror
- Fixed event listener cleanup to prevent accumulation
- Added comprehensive test coverage for memory leak prevention
This fixes the issue where MCP server would lose connection after
several hours due to timer accumulation causing memory exhaustion.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add FixedCollectionValidator utility to handle all fixedCollection patterns
- Support validation for 12 different node types including Switch, If, Filter,
Summarize, Compare Datasets, Sort, Aggregate, Set, HTML, HTTP Request, and Airtable
- Refactor enhanced-config-validator to use the generic utility
- Add comprehensive tests with 19 test cases covering all node types
- Maintain backward compatibility with existing validation behavior
This prevents the 'propertyValues[itemName] is not iterable' error across all
susceptible n8n nodes, not just Switch/If/Filter.
- Add validation for invalid fixedCollection structures in Switch, If, and Filter nodes
- Detect and prevent nested 'conditions.values' patterns that cause n8n UI crashes
- Support both 'n8n-nodes-base.x' and 'nodes-base.x' node type formats
- Provide auto-fix suggestions for invalid structures
- Add comprehensive test coverage for all edge cases
This prevents AI agents from creating invalid node configurations that break n8n's UI.
- Add validateToolParams method with clear error messages
- Fix failing tests to expect new parameter validation errors
- Create comprehensive parameter validation test suite (52 tests)
- Add parameter validation for all n8n management tools
- Test numeric parameter conversion and edge cases
- Ensure consistent error format across all tools
- Verify MCP error response handling
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Fix express.json() mocking issue in tests by properly creating express mock
- Update test expectations to match new security-enhanced response format
- Adjust CORS test to include DELETE method added for session management
- All n8n mode tests now passing with security features intact
The server now includes:
- Production token validation with minimum 32 character requirement
- Session limiting (max 100 concurrent sessions)
- Automatic session cleanup every 5 minutes
- Enhanced health endpoint with security and session metrics
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add N8N_MODE environment variable for n8n-specific behavior
- Implement HTTP Streamable transport with multiple session support
- Add protocol version endpoint (GET /mcp) for n8n compatibility
- Support multiple initialize requests for stateless n8n clients
- Add Docker configuration for n8n deployment
- Add test script with persistent volume support
- Add comprehensive unit tests for n8n mode
- Fix session management to handle per-request transport pattern
BREAKING CHANGE: Server now creates new transport for each initialize request
when running in n8n mode to support n8n's stateless client architecture
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Major improvements based on comprehensive test suite review:
Test Fixes:
- Fix all 78 failing tests across logger, MSW, and validator tests
- Fix console spy management in logger tests with proper DEBUG env handling
- Fix MSW test environment restoration in session-management.test.ts
- Fix workflow validator tests by adding proper node connections
- Fix mock setup issues in edge case tests
Test Organization:
- Split large config-validator.test.ts (1,075 lines) into 4 focused files
- Rename 63+ tests to follow "should X when Y" naming convention
- Add comprehensive edge case test files for all major validators
- Create tests/README.md with testing guidelines and best practices
New Features:
- Add ConfigValidator.validateBatch() method for bulk validation
- Add edge case coverage for null/undefined, boundaries, invalid data
- Add CI-aware performance test timeouts
- Add JSDoc comments to test utilities and factories
- Add workflow duplicate node name validation tests
Results:
- All tests passing: 1,356 passed, 19 skipped
- Test coverage: 85.34% statements, 85.3% branches
- From 78 failures to 0 failures
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove AI agent coordination files and progress tracking
- Remove temporary test results and generated artifacts
- Remove diagnostic test scripts from src/scripts/
- Remove development planning documents
- Update .gitignore to exclude test artifacts
- Clean up 53 temporary files total
- Fixed undefined variable reference in server.ts (possiblePaths)
- Fixed type mismatches in database performance tests
- Added proper type assertions for MCP response objects
- Fixed TemplateNode interface compliance in tests
All TypeScript checks now pass successfully.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Fixed better-sqlite3 ES module imports across all tests
- Updated template repository method to handle undefined results
- Fixed all database column references to match schema
- Corrected MCP transport initialization
- All integration tests now passing
- Create benchmark test suites for critical operations:
- Node loading performance
- Database query performance
- Search operations performance
- Validation performance
- MCP tool execution performance
- Add GitHub Actions workflow for benchmark tracking:
- Runs on push to main and PRs
- Uses github-action-benchmark for historical tracking
- Comments on PRs with performance results
- Alerts on >10% performance regressions
- Stores results in GitHub Pages
- Create benchmark infrastructure:
- Custom Vitest benchmark configuration
- JSON reporter for CI results
- Result formatter for github-action-benchmark
- Performance threshold documentation
- Add supporting utilities:
- SQLiteStorageService for benchmark database setup
- MCPEngine wrapper for testing MCP tools
- Test factories for generating benchmark data
- Enhanced NodeRepository with benchmark methods
- Document benchmark system:
- Comprehensive benchmark guide in docs/BENCHMARKS.md
- Performance thresholds in .github/BENCHMARK_THRESHOLDS.md
- README for benchmarks directory
- Integration with existing test suite
The benchmark system will help monitor performance over time and catch regressions before they reach production.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Create comprehensive test suite with 69 tests for WorkflowValidator
- Increase coverage from 2.32% to 97.59%
- Fix bugs in WorkflowValidator:
- Add null checks for workflow.nodes before accessing length
- Fix checkNodeErrorHandling to process each node individually
- Fix disabled node validation logic
- Ensure error-prone nodes generate proper warnings
- Test all major methods and edge cases:
- validateWorkflow with different options
- validateAllNodes with various node types
- validateConnections including cycles and orphans
- validateExpressions with complex expressions
- checkWorkflowPatterns for best practices
- checkNodeErrorHandling for error configurations
- generateSuggestions for helpful tips
All 69 tests now passing with excellent coverage.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove Jest and all related packages
- Install Vitest with coverage support
- Create vitest.config.ts with path aliases
- Set up global test configuration
- Migrate all 6 test files to Vitest syntax
- Update TypeScript configuration for better Vitest support
- Create separate tsconfig.build.json for clean builds
- Fix all import/module issues in tests
- All 68 tests passing successfully
- Current coverage baseline: 2.45%
Phase 1 of testing suite improvement complete.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Added proper SIGTERM/SIGINT signal handlers to stdio-wrapper.ts
- Removed problematic trap commands from docker-entrypoint.sh
- Added STOPSIGNAL directive to Dockerfile for explicit signal handling
- Implemented graceful shutdown in MCP server with database cleanup
- Added stdin close detection for proper cleanup when Claude Desktop closes the pipe
- Containers now properly exit with the --rm flag, preventing accumulation
- Added --init flag to all Docker configuration examples
- Updated documentation with container lifecycle management best practices
- Bumped version to 2.7.20
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Added support for n8n-nodes-langchain.* → nodes-langchain.* normalization
- Implemented case-insensitive node name matching (e.g., chattrigger → chatTrigger)
- Added smart camelCase detection for common patterns (trigger, request, sheets, etc.)
- Fixed get_node_documentation tool to use same normalization logic as other tools
- Updated all 7 node lookup locations to use normalized types for alternatives
- Enhanced getNodeTypeAlternatives() to normalize all generated alternatives
All MCP tools now consistently handle various format variations:
- nodes-langchain.chatTrigger (correct format)
- n8n-nodes-langchain.chatTrigger (package format)
- n8n-nodes-langchain.chattrigger (package + wrong case)
- nodes-langchain.chattrigger (wrong case only)
- @n8n/n8n-nodes-langchain.chatTrigger (full npm format)
Bump version to 2.7.19
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Changed misleading 'total' field to 'returned' to clarify it's the count in current page
- Added 'hasMore' boolean flag for clear pagination indication
- Added '_note' guidance when more data is available
- Applied same improvements to n8n_list_executions for consistency
Performance improvements:
- Tool now returns only minimal metadata instead of full workflow structure
- Reduced response size by ~95% (from thousands to ~10 tokens per workflow)
- Eliminated token limit errors when listing workflows with many nodes
- Updated descriptions and documentation to clarify minimal response
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Fixed health check to use correct /healthz endpoint instead of /health
- Added MCP version (mcpVersion) and supported n8n version (supportedN8nVersion) to health check response
- Added versionNote field with instructions for AI agents about manual version verification
- n8n API limitation: instance version cannot be determined automatically
- Updated axios usage for healthz endpoint access with proper error handling
- Added workflowNodeType field to all node-returning MCP tools
- AI agents now receive both internal format (nodes-base.webhook) and workflow format (n8n-nodes-base.webhook)
- Created getWorkflowNodeType() utility to construct proper n8n format from package name
- Solves issue where AI agents would search nodes and use wrong format in workflows
- No database changes required - uses existing package_name field
- Updated: search_nodes, get_node_info, get_node_essentials, get_node_as_tool_info, validate_node_operation
- Updated CHANGELOG.md with comprehensive documentation of the changes
This completes the fix for issue #71, ensuring AI agents can seamlessly create workflows
with the correct node type format without manual intervention.
- Add centralized normalizeNodeType utility to handle prefix conversion
- n8n-nodes-base.* → nodes-base.*
- @n8n/n8n-nodes-langchain.* → nodes-langchain.*
- Update all 9 affected MCP tools to use normalized node types
- AI agents can now use node types directly from n8n workflow exports
- Maintains backward compatibility with existing shortened prefixes
- Add comprehensive test coverage for all affected methods
Fixes#71🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove examples from get_node_essentials responses
- Remove examples from validate_node_operation when errors occur
- Update documentation to reflect removal of examples
- Keep helpful format hints in get_node_for_task (different purpose)
The auto-generated examples were misleading AI agents with incorrect
configurations (e.g., Slack "channel" vs "select" property). Tools
now focus on validation and error messages instead of examples.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Redesigned documentation to be utilitarian and AI-agent focused
- Removed all pleasantries, emojis, and conversational language
- Added concrete numbers throughout (528 nodes, 108 triggers, 264 AI tools)
- Updated all tool descriptions with practical, actionable information
- Enhanced examples with actual return structures and usage patterns
- Made Code node guides prominently featured in overview
- Verified documentation accuracy through extensive testing
- Standardized format across all 30+ tool documentation files
Documentation now optimized for token efficiency while maintaining
clarity and completeness for AI agent consumption.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Migrated all 40 MCP tools documentation to modular structure
- Created comprehensive documentation with both essentials and full details
- Organized tools by category: discovery, configuration, validation, templates, workflow_management, system, special
- Fixed all TODO placeholders with informative, precise content
- Each tool now has concise description, key tips, and full documentation
- Improved documentation quality: 30-40% more concise while maintaining usefulness
- Fixed TypeScript compilation issues and removed orphaned content
- All tools accessible via tools_documentation MCP endpoint
- Build successful with zero errors
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Fixed docker-entrypoint.sh to use NODE_DB_PATH instead of hardcoded paths
- Added log_message() helper to prevent stdio mode output corruption
- Fixed directory creation race condition with proper ownership
- Added path validation to ensure NODE_DB_PATH ends with .db
- Updated rebuild.ts to respect NODE_DB_PATH environment variable
- Added comprehensive documentation for custom database paths
- Bumped version to 2.7.16
The bug was caused by hardcoded /app/data/nodes.db paths in the Docker
entrypoint script that ignored the NODE_DB_PATH environment variable.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add intelligent URL detection supporting BASE_URL, PUBLIC_URL, and proxy headers
- Fix hardcoded localhost URLs in server console output
- Add hostname validation to prevent host header injection attacks
- Restrict URL schemes to http/https only (block javascript:, file://, etc.)
- Remove sensitive environment data from API responses
- Add GET endpoints (/, /mcp) for better API discovery
- Fix version inconsistency between server implementations
- Update HTTP bridge to use HOST/PORT environment variables
- Add comprehensive test scripts for URL configuration and security
This resolves issues #41 and #42 by making the HTTP server properly handle
deployment behind reverse proxies and adds critical security validations.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove default settings logic from cleanWorkflowForUpdate that was causing
"settings must NOT have additional properties" error
- The function now only removes read-only fields without adding any properties
- Add comprehensive test coverage in test-issue-45-fix.ts
- Add documentation explaining the difference between create and update functions
- Bump version to 2.7.14
This fixes the issue where n8n_update_partial_workflow would pass validation
but fail during execution when workflows didn't have settings defined.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Enhanced database adapter to support multiple WASM file resolution strategies
- Added require.resolve() for reliable package location in npm environments
- Made better-sqlite3 an optional dependency
- Improved error handling with clear messages
- Updated version to 2.7.13
- Updated CHANGELOG and README badges
- Database now contains 499 templates (was 399)
- FTS5 index properly populated with all template entries
- Fixed quote escaping in FTS5 queries to prevent syntax errors
- Verified FTS5 search returns correct results (162 for "webhook")
- Fixes template search in Docker deployments
The previous database had empty FTS5 tables causing search to fail.
This update ensures the FTS5 index is properly synchronized and
handles special characters in search queries.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add FTS5 pre-creation in fetch-templates.ts before data import
- Create prebuild-fts5.ts script to ensure FTS5 tables exist
- Improve logging in template-repository.ts for better debugging
- Add npm script 'prebuild:fts5' for manual FTS5 setup
This ensures template search works consistently in Docker mode
where runtime FTS5 table creation might fail due to permissions.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Added runtime FTS5 detection in database adapters
- Removed FTS5 from required schema to prevent "no such module" errors
- FTS5 tables/triggers created conditionally only if supported
- Template search automatically falls back to LIKE when FTS5 unavailable
- Works in ALL SQLite environments (Claude Desktop, restricted envs, etc.)
This ensures search_templates() works correctly regardless of SQLite build,
while still providing optimal performance when FTS5 is available.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Reduced average description length from 250-450 to 93-129 chars
- Documentation tools now average 129 chars per description
- Management tools average just 93 chars per description
- Moved detailed documentation to tools_documentation() system
- Only 2 tools exceed 200 chars (necessarily verbose)
Also includes search_nodes improvements:
- Fixed primary node ranking (webhook, HTTP Request now appear first)
- Fixed FUZZY mode threshold for better typo tolerance
- Removed unnecessary searchInfo messages
- Fixed HTTP node type case sensitivity issue
This significantly improves AI agent performance by reducing context usage
while preserving all essential information.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Critical fixes based on Claude Desktop feedback:
1. Fixed crypto documentation: require('crypto') IS available despite editor warnings
- Added clear examples of crypto usage
- Updated validation to guide correct require() usage
2. Clarified $helpers vs standalone functions
- $getWorkflowStaticData() is standalone, NOT $helpers.getWorkflowStaticData()
- Added validation to catch incorrect usage (prevents '$helpers is not defined' errors)
- Enhanced examples showing proper $helpers availability checks
3. Fixed JMESPath numeric literal documentation
- n8n requires backticks around numbers in filters: [?age >= `18`]
- Added multiple examples and validation to detect missing backticks
- Prevents 'JMESPath syntax error' that Claude Desktop encountered
4. Fixed webhook data access gotcha
- Webhook payload is at items[0].json.body, NOT items[0].json
- Added dedicated 'Webhook Data Access' section with clear examples
- Created process_webhook_data task template
- Added validation to detect incorrect webhook data access patterns
All fixes based on production workflows TaNqYoZNNeHC4Hne and JZ9urD7PNClDZ1bm
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Root cause: AI agents were placing error handling properties inside `parameters` instead of at node level
Major changes:
- Enhanced workflow validator to check for ALL node-level properties (expanded from 6 to 11)
- Added validation for onError property values and deprecation warnings for continueOnFail
- Updated all examples to use modern error handling (onError instead of continueOnFail)
- Added comprehensive node-level properties documentation in tools_documentation
- Enhanced MCP tool documentation for n8n_create_workflow and n8n_update_partial_workflow
- Added test script demonstrating correct node-level property usage
Node-level properties now validated:
- credentials, disabled, notes, notesInFlow, executeOnce
- onError, retryOnFail, maxTries, waitBetweenTries, alwaysOutputData
- continueOnFail (deprecated)
Validation improvements:
- Detects misplaced properties and provides clear fix examples
- Shows complete node structure when properties are incorrectly placed
- Type validation for all node-level boolean and string properties
- Smart error messages with correct placement guidance
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Added specific error reasons for auth failures: no_auth_header, invalid_auth_format, invalid_token
- Fixed AUTH_TOKEN_FILE support in Docker production stacks (issue #16)
- Added AUTH_TOKEN_FILE support to single-session HTTP server for consistency
- Enhanced security by removing token lengths from logs
- Added token trimming and empty token validation
- Updated Docker entrypoint to properly support AUTH_TOKEN_FILE
- Bumped version to 2.7.10
This improves debugging for mcp-remote authentication issues and enables
proper Docker secrets usage in production environments.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
