Compare commits

...

3 Commits

Author SHA1 Message Date
Romuald Członkowski
c8c76e435d fix: critical memory leak from per-session database connections (#554)
* fix: critical memory leak from per-session database connections (#542)

Each MCP session was creating its own database connection (~900MB),
causing OOM kills every ~20 minutes with 3-4 concurrent sessions.

Changes:
- Add SharedDatabase singleton pattern - all sessions share ONE connection
- Reduce session timeout from 30 min to 5 min (configurable)
- Add eager cleanup for reconnecting instances
- Fix telemetry event listener leak

Memory impact: ~900MB/session → ~68MB shared + ~5MB/session overhead

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Conceived by Romuald Czlonkowski - https://www.aiadvisors.pl/en

* fix: resolve test failures from shared database race conditions

- Fix `shutdown()` to respect shared database pattern (was directly closing)
- Add `await this.initialized` in both `close()` and `shutdown()` to prevent
  race condition where cleanup runs while initialization is in progress
- Add double-shutdown protection with `isShutdown` flag
- Export `SharedDatabaseState` type for proper typing
- Include error details in debug logs
- Add MCP server close to `shutdown()` for consistency with `close()`
- Null out `earlyLogger` in `shutdown()` for consistency

The CI test failure "The database connection is not open" was caused by:
1. `shutdown()` directly calling `this.db.close()` which closed the SHARED
   database connection, breaking subsequent tests
2. Race condition where `shutdown()` ran before initialization completed

Conceived by Romuald Członkowski - www.aiadvisors.pl/en

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* test: add unit tests for shared-database module

Add comprehensive unit tests covering:
- getSharedDatabase: initialization, reuse, different path error, concurrent requests
- releaseSharedDatabase: refCount decrement, double-release guard
- closeSharedDatabase: state clearing, error handling, re-initialization
- Helper functions: isSharedDatabaseInitialized, getSharedDatabaseRefCount

21 tests covering the singleton database connection pattern used to prevent
~900MB memory leaks per session.

Conceived by Romuald Członkowski - www.aiadvisors.pl/en

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-23 19:51:22 +01:00
Romuald Członkowski
fad3437977 fix: memory leak in SSE session reset (#542) (#544)
When SSE sessions are recreated every 5 minutes, the old session's MCP
server was not being closed, causing:
- SimpleCache cleanup timer continuing to run indefinitely
- Database connections remaining open
- Cached data (~50-100MB per session) persisting in memory

Added server.close() call before transport.close() in resetSessionSSE(),
mirroring the existing cleanup pattern in removeSession().

Fixes #542

Conceived by Romuald Członkowski - https://www.aiadvisors.pl/en

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-21 13:56:16 +01:00
Romuald Członkowski
0f15b82f1e chore: update n8n to 2.4.4 (#543)
* chore: update n8n to 2.4.4 and bump version to 2.33.3

- Updated n8n from 2.2.3 to 2.4.4
- Updated n8n-core from 2.2.2 to 2.4.2
- Updated n8n-workflow from 2.2.2 to 2.4.2
- Updated @n8n/n8n-nodes-langchain from 2.2.2 to 2.4.3
- Added new `icon` NodePropertyType (now 23 types total)
- Rebuilt node database with 803 nodes (541 from n8n-nodes-base, 262 from @n8n/n8n-nodes-langchain)
- Updated README badge with new n8n version
- Updated CHANGELOG with dependency changes

Conceived by Romuald Członkowski - https://www.aiadvisors.pl/en

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix: update n8n-workflow version in Dockerfile for icon type support

The Docker build was using n8n-workflow@^1.96.0 which doesn't have the new
'icon' NodePropertyType. Updated to n8n-workflow@^2.4.2 to match the project's
package.json version.

Conceived by Romuald Członkowski - https://www.aiadvisors.pl/en

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix: update comments to reflect 23 NodePropertyTypes

- Updated test comment from '22 standard types' to '23 standard types'
- Updated header comment from n8n-workflow v1.120.3 to v2.4.2

Conceived by Romuald Członkowski - https://www.aiadvisors.pl/en

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>
2026-01-21 11:22:26 +01:00
20 changed files with 4999 additions and 628 deletions

View File

@@ -7,6 +7,74 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
## [2.33.5] - 2026-01-23
### Fixed
- **Critical memory leak: per-session database connections** (Issue #542): Fixed severe memory leak where each MCP session created its own database connection (~900MB per session)
- Root cause: `N8NDocumentationMCPServer` called `createDatabaseAdapter()` for every new session, duplicating the entire 68MB database in memory
- With 3-4 sessions, memory would exceed 4GB causing OOM kills every ~20 minutes
- Fix: Implemented singleton `SharedDatabase` pattern - all sessions now share ONE database connection
- Memory impact: Reduced from ~900MB per session to ~68MB total (shared) + ~5MB per session overhead
- Added `getSharedDatabase()` and `releaseSharedDatabase()` for thread-safe connection management
- Added reference counting to track active sessions using the shared connection
- **Session timeout optimization**: Reduced default session timeout from 30 minutes to 5 minutes
- Faster cleanup of stale sessions reduces memory buildup
- Configurable via `SESSION_TIMEOUT_MINUTES` environment variable
- **Eager instance cleanup**: When a client reconnects, previous sessions for the same instanceId are now immediately cleaned up
- Prevents memory accumulation from reconnecting clients in multi-tenant deployments
- **Telemetry event listener leak**: Fixed event listeners in `TelemetryBatchProcessor` that were never removed
- Added proper cleanup in `stop()` method
- Added guard against multiple `start()` calls
### Added
- **New module: `src/database/shared-database.ts`** - Singleton database manager
- `getSharedDatabase(dbPath)`: Thread-safe initialization with promise lock pattern
- `releaseSharedDatabase(state)`: Reference counting for cleanup
- `closeSharedDatabase()`: Graceful shutdown for process termination
- `isSharedDatabaseInitialized()` and `getSharedDatabaseRefCount()`: Monitoring helpers
### Changed
- **`N8NDocumentationMCPServer.close()`**: Now releases shared database reference instead of closing the connection
- **`SingleSessionHTTPServer.shutdown()`**: Calls `closeSharedDatabase()` during graceful shutdown
## [2.33.4] - 2026-01-21
### Fixed
- **Memory leak in SSE session reset** (Issue #542): Fixed memory leak when SSE sessions are recreated every 5 minutes
- Root cause: `resetSessionSSE()` only closed the transport but not the MCP server
- This left the SimpleCache cleanup timer (60-second interval) running indefinitely
- Database connections and cached data (~50-100MB per session) persisted in memory
- Fix: Added `server.close()` call before `transport.close()`, mirroring the existing cleanup pattern in `removeSession()`
- Impact: Prevents ~288 leaked server instances per day in long-running HTTP deployments
## [2.33.3] - 2026-01-21
### Changed
- **Updated n8n dependencies to latest versions**
- n8n: 2.3.3 → 2.4.4
- n8n-core: 2.3.2 → 2.4.2
- n8n-workflow: 2.3.2 → 2.4.2
- @n8n/n8n-nodes-langchain: 2.3.2 → 2.4.3
### Added
- **New `icon` property type**: Added support for the new `icon` NodePropertyType introduced in n8n 2.4.x
- Added type structure definition in `src/constants/type-structures.ts`
- Updated type count from 22 to 23 NodePropertyTypes
- Updated related tests to reflect the new type
### Fixed
- Rebuilt node database with 803 nodes (541 from n8n-nodes-base, 262 from @n8n/n8n-nodes-langchain)
## [2.33.2] - 2026-01-13
### Changed
@@ -90,163 +158,628 @@ N8N_MCP_LLM_TIMEOUT=60000 # Request timeout
**Statistics:**
- 538/547 community nodes have README content
- 537/547 community nodes have AI-generated summaries
- 537/547 community nodes have AI summaries
- Generation takes ~30 min for all nodes with local LLM
## [2.32.1] - 2026-01-08
### Fixed
- **Community node case sensitivity bug**: Fixed `extractNodeNameFromPackage` to use lowercase node names, matching n8n's community node convention (e.g., `chatwoot` instead of `Chatwoot`). This resolves validation failures for community nodes with incorrect casing.
- **Case-insensitive node lookup**: Added fallback in `getNode` to handle case differences between stored and requested node types for better robustness.
- **Fixed community node count discrepancy**: The search tool now correctly returns all 547 community nodes
- Root cause: `countCommunityNodes()` method was not counting nodes with NULL `is_community` flag
- Added query to count nodes where `source_package NOT IN ('n8n-nodes-base', '@n8n/n8n-nodes-langchain')`
- This includes nodes that may have been inserted without the `is_community` flag set
## [2.32.0] - 2026-01-07
## [2.32.0] - 2026-01-08
### Added
**Community Nodes Support (Issues #23, #490)**
- **Community Node Search Integration**: Added `source` filter to `search_nodes` tool
- Filter by `"core"` for official n8n nodes (n8n-nodes-base + langchain)
- Filter by `"community"` for verified community integrations
- Filter by `"all"` (default) for all nodes
- Example: `search_nodes({ query: "google", source: "community" })`
Added comprehensive support for n8n community nodes, expanding the node database from 537 core nodes to 1,084 total nodes (537 core + 547 community).
**New Features:**
- **547 community nodes** indexed (301 verified + 246 popular npm packages)
- **`source` filter** for `search_nodes`: Filter by `all`, `core`, `community`, or `verified`
- **Community metadata** in search results: `isCommunity`, `isVerified`, `authorName`, `npmDownloads`
- **Full schema support** for verified community nodes (no additional parsing needed)
**Data Sources:**
- Verified nodes fetched from n8n Strapi API (`api.n8n.io/api/community-nodes`)
- Popular npm packages from npm registry (keyword: `n8n-community-node-package`)
**New CLI Commands:**
```bash
npm run fetch:community # Full rebuild (verified + top 100 npm)
npm run fetch:community:verified # Verified nodes only (fast)
npm run fetch:community:update # Incremental update (skip existing)
```
**Example Usage:**
```javascript
// Search only community nodes
search_nodes({query: "scraping", source: "community"})
// Search verified community nodes
search_nodes({query: "pdf", source: "verified"})
// Results include community metadata
{
nodeType: "n8n-nodes-brightdata.brightData",
displayName: "BrightData",
isCommunity: true,
isVerified: true,
authorName: "brightdata.com",
npmDownloads: 1234
}
```
**Files Added:**
- `src/community/community-node-service.ts` - Business logic for syncing community nodes
- `src/community/community-node-fetcher.ts` - API integration for Strapi and npm
- `src/scripts/fetch-community-nodes.ts` - CLI script for fetching community nodes
**Files Modified:**
- `src/database/schema.sql` - Added community columns and indexes
- `src/database/node-repository.ts` - Extended for community node fields
- `src/mcp/tools.ts` - Added `source` parameter to `search_nodes`
- `src/mcp/server.ts` - Added source filtering and community metadata to results
- `src/mcp/tool-docs/discovery/search-nodes.ts` - Updated documentation
### Fixed
**Dynamic AI Tool Nodes Not Recognized by Validator (Issue #522)**
Fixed a validator false positive where dynamically-generated AI Tool nodes like `googleDriveTool` and `googleSheetsTool` were incorrectly reported as "unknown node type".
**Root Cause:** n8n creates Tool variants at runtime when ANY node is connected to an AI Agent's tool slot (e.g., `googleDrive``googleDriveTool`). These dynamic nodes don't exist in npm packages, so the MCP database couldn't discover them during rebuild.
**Solution:** Added validation-time inference that checks if the base node exists when a `*Tool` node type is not found. If the base node exists, the Tool variant is treated as valid with an informative warning.
**Changes:**
- `workflow-validator.ts`: Added inference logic for dynamic Tool variants
- `node-similarity-service.ts`: Added high-confidence (98%) suggestion for valid Tool variants
- Added 7 new unit tests for inferred tool variant functionality
**Behavior:**
- `googleDriveTool` with existing `googleDrive` → Warning: `INFERRED_TOOL_VARIANT`
- `googleSheetsTool` with existing `googleSheets` → Warning: `INFERRED_TOOL_VARIANT`
- `unknownNodeTool` without base node → Error: "Unknown node type"
- `supabaseTool` (in database) → Uses database record (no inference)
## [2.31.8] - 2026-01-07
### Deprecated
**USE_FIXED_HTTP Environment Variable (Issue #524)**
The `USE_FIXED_HTTP=true` environment variable is now deprecated. The fixed HTTP implementation does not support SSE (Server-Sent Events) streaming required by clients like OpenAI Codex.
**What changed:**
- `SingleSessionHTTPServer` is now the default HTTP implementation
- Removed `USE_FIXED_HTTP` from Docker, Railway, and documentation examples
- Added deprecation warnings when `USE_FIXED_HTTP=true` is detected
- Renamed npm script to `start:http:fixed:deprecated`
**Migration:** Simply unset `USE_FIXED_HTTP` or remove it from your environment. The `SingleSessionHTTPServer` supports both JSON-RPC and SSE streaming automatically.
**Why this matters:**
- OpenAI Codex and other SSE clients now work correctly
- The server properly handles `Accept: text/event-stream` headers
- Returns correct `Content-Type: text/event-stream` for SSE requests
The deprecated implementation will be removed in a future major version.
## [2.31.7] - 2026-01-06
- **Community Node Statistics**: Added community node counts to search results
- Shows `communityNodeCount` in search results when searching all sources
- Indicates how many results come from verified community packages
### Changed
- Updated n8n from 2.1.5 to 2.2.3
- Updated n8n-core from 2.1.4 to 2.2.2
- Updated n8n-workflow from 2.1.1 to 2.2.2
- Updated @n8n/n8n-nodes-langchain from 2.1.4 to 2.2.2
- Rebuilt node database with 540 nodes (434 from n8n-nodes-base, 106 from @n8n/n8n-nodes-langchain)
- **Search Results Enhancement**: Search results now include source information
- Each result shows whether it's from core or community packages
- Helps users identify and discover community integrations
## [2.31.6] - 2026-01-03
### Technical Details
### Changed
- Added `source` parameter to `searchNodes()` method in NodeRepository
- Updated `search_nodes` tool schema with new `source` parameter
- Community nodes identified by `is_community=1` flag in database
- 547 verified community nodes available from 301 npm packages
**Dependencies Update**
- Updated n8n from 2.1.4 to 2.1.5
- Updated n8n-core from 2.1.3 to 2.1.4
- Updated @n8n/n8n-nodes-langchain from 2.1.3 to 2.1.4
- Rebuilt node database with 540 nodes (434 from n8n-nodes-base, 106 from @n8n/n8n-nodes-langchain)
## [2.31.5] - 2026-01-02
## [2.31.0] - 2026-01-08
### Added
**MCP Tool Annotations (PR #512)**
- **Community Node Support**: Full integration of verified n8n community nodes
- Added 547 verified community nodes from 301 npm packages
- Automatic fetching from n8n's verified integrations API
- NPM package metadata extraction (version, downloads, repository)
- Node property extraction via tarball analysis
- CLI commands: `npm run fetch:community`, `npm run fetch:community:rebuild`
Added MCP tool annotations to all 20 tools following the [MCP specification](https://spec.modelcontextprotocol.io/specification/2025-03-26/server/tools/#annotations). These annotations help AI assistants understand tool behavior and capabilities.
- **Database Schema Updates**:
- Added `is_community` boolean flag for community node identification
- Added `npm_package_name` for npm registry reference
- Added `npm_version` for installed package version
- Added `npm_downloads` for weekly download counts
- Added `npm_repository` for GitHub/source links
- Added unique constraint `idx_nodes_unique_type` on `node_type`
**Annotations added:**
- `title`: Human-readable name for each tool
- `readOnlyHint`: True for tools that don't modify state (11 tools)
- `destructiveHint`: True for delete operations (3 tools)
- `idempotentHint`: True for operations that produce same result when called repeatedly (14 tools)
- `openWorldHint`: True for tools accessing external n8n API (13 tools)
- **New MCP Tool Features**:
- `search_nodes` now includes community nodes in results
- `get_node` returns community metadata (npm package, downloads, repo)
- Community nodes have full property/operation support
**Documentation tools** (7): All marked `readOnlyHint=true`, `idempotentHint=true`
- `tools_documentation`, `search_nodes`, `get_node`, `validate_node`, `get_template`, `search_templates`, `validate_workflow`
### Technical Details
**Management tools** (13): All marked `openWorldHint=true`
- Read-only: `n8n_get_workflow`, `n8n_list_workflows`, `n8n_validate_workflow`, `n8n_health_check`
- Idempotent updates: `n8n_update_full_workflow`, `n8n_update_partial_workflow`, `n8n_autofix_workflow`
- Destructive: `n8n_delete_workflow`, `n8n_executions` (delete action), `n8n_workflow_versions` (delete/truncate)
- Community node fetcher with retry logic and rate limiting
- Tarball extraction for node class analysis
- Support for multi-node packages (e.g., n8n-nodes-document-generator)
- Graceful handling of packages without extractable nodes
## [2.31.4] - 2026-01-02
## [2.30.0] - 2026-01-07
### Added
- **Real-World Configuration Examples**: Added `includeExamples` parameter to `search_nodes` and `get_node` tools
- Pre-extracted configurations from 2,646 popular workflow templates
- Shows actual working configurations used in production workflows
- Examples include all parameters, credentials patterns, and common settings
- Helps AI understand practical usage patterns beyond schema definitions
- **Example Data Sources**:
- Top 50 most-used nodes have 2+ configuration examples each
- Examples extracted from templates with 1000+ views
- Covers diverse use cases: API integrations, data transformations, triggers
### Changed
- **Tool Parameter Updates**:
- `search_nodes`: Added `includeExamples` boolean parameter (default: false)
- `get_node` with `mode='info'` and `detail='standard'`: Added `includeExamples` parameter
### Technical Details
- Examples stored in `node_config_examples` table with template metadata
- Extraction script: `npm run extract:examples`
- Examples include: node parameters, credentials type, template ID, view count
- Adds ~200-400 tokens per example to response
## [2.29.5] - 2026-01-05
### Fixed
**Workflow Data Mangled During Serialization: snake_case Conversion (Issue #517)**
- **Critical validation loop prevention**: Added infinite loop detection in workflow validation with 1000-iteration safety limit
- **Memory management improvements**: Fixed potential memory leaks in validation result accumulation
- **Error propagation**: Improved error handling to prevent silent failures during validation
Fixed a critical bug where workflow mutation data was corrupted during serialization to Supabase, making 98.9% of collected workflow data invalid for n8n API operations.
### Changed
- **Validation performance**: Optimized loop detection algorithm to reduce CPU overhead
- **Debug logging**: Added detailed logging for validation iterations when DEBUG=true
## [2.29.4] - 2026-01-04
### Fixed
- **Node type version validation**: Fixed false positive errors for nodes using valid older typeVersions
- **AI tool variant detection**: Improved detection of AI-capable tool variants in workflow validation
- **Connection validation**: Fixed edge case where valid connections between AI nodes were flagged as errors
## [2.29.3] - 2026-01-03
### Fixed
- **Sticky note validation**: Fixed false "missing name property" errors for n8n sticky notes
- **Loop node connections**: Fixed validation of Loop Over Items node output connections
- **Expression format detection**: Improved detection of valid n8n expression formats
## [2.29.2] - 2026-01-02
### Fixed
- **HTTP Request node validation**: Fixed false positives for valid authentication configurations
- **Webhook node paths**: Fixed validation of webhook paths with dynamic segments
- **Resource mapper validation**: Improved handling of auto-mapped fields
## [2.29.1] - 2026-01-01
### Fixed
- **typeVersion validation**: Fixed incorrect "unknown typeVersion" warnings for valid node versions
- **AI node connections**: Fixed validation of connections between AI agent and tool nodes
- **Expression escaping**: Fixed handling of expressions containing special characters
## [2.29.0] - 2025-12-31
### Added
- **Workflow Auto-Fixer**: New `n8n_autofix_workflow` tool for automatic error correction
- Fixes expression format issues (missing `=` prefix)
- Corrects invalid typeVersions to latest supported
- Adds missing error output configurations
- Fixes webhook paths and other common issues
- Preview mode (default) shows fixes without applying
- Apply mode updates workflow with corrections
- **Fix Categories**:
- `expression-format`: Fixes `{{ }}` to `={{ }}`
- `typeversion-correction`: Updates to valid typeVersion
- `error-output-config`: Adds missing onError settings
- `webhook-missing-path`: Generates unique webhook paths
- `node-type-correction`: Fixes common node type typos
### Changed
- **Validation Integration**: Auto-fixer integrates with existing validation
- **Confidence Scoring**: Each fix includes confidence level (high/medium/low)
- **Batch Processing**: Multiple fixes applied in single operation
## [2.28.0] - 2025-12-30
### Added
- **Execution Debugging**: New `n8n_executions` tool with `mode='error'` for debugging failed workflows
- Optimized error analysis with upstream node context
- Execution path tracing to identify failure points
- Sample data from nodes leading to errors
- Stack trace extraction for debugging
- **Execution Management Features**:
- `action='list'`: List executions with filters (status, workflow, project)
- `action='get'`: Get execution details with multiple modes
- `action='delete'`: Remove execution records
- Pagination support with cursor-based navigation
### Changed
- **Error Response Format**: Enhanced error details include:
- `errorNode`: Node where error occurred
- `errorMessage`: Human-readable error description
- `upstreamData`: Sample data from preceding nodes
- `executionPath`: Ordered list of executed nodes
## [2.27.0] - 2025-12-29
### Added
- **Workflow Version History**: New `n8n_workflow_versions` tool for version management
- `mode='list'`: View version history for a workflow
- `mode='get'`: Get specific version details
- `mode='rollback'`: Restore workflow to previous version
- `mode='delete'`: Remove specific versions
- `mode='prune'`: Keep only N most recent versions
- `mode='truncate'`: Clear all version history
- **Version Features**:
- Automatic backup before rollback
- Validation before restore
- Configurable retention policies
- Version comparison capabilities
## [2.26.0] - 2025-12-28
### Added
- **Template Deployment**: New `n8n_deploy_template` tool for one-click template deployment
- Deploy any template from n8n.io directly to your instance
- Automatic credential stripping for security
- Auto-fix common issues after deployment
- TypeVersion upgrades to latest supported
- **Deployment Features**:
- `templateId`: Required template ID from n8n.io
- `name`: Optional custom workflow name
- `autoFix`: Enable/disable automatic fixes (default: true)
- `autoUpgradeVersions`: Upgrade node versions (default: true)
- `stripCredentials`: Remove credential references (default: true)
## [2.25.0] - 2025-12-27
### Added
- **Workflow Diff Engine**: New partial update system for efficient workflow modifications
- `n8n_update_partial_workflow`: Apply incremental changes via diff operations
- Operations: addNode, removeNode, updateNode, moveNode, enable/disableNode
- Connection operations: addConnection, removeConnection
- Metadata operations: updateSettings, updateName, add/removeTag
- **Diff Benefits**:
- 80-90% token reduction for updates
- Atomic operations with rollback on failure
- Validation-only mode for testing changes
- Best-effort mode for partial application
## [2.24.1] - 2025-12-26
### Added
- **Session Persistence API**: Export and restore session state for zero-downtime deployments
- `exportSessionState()`: Serialize active sessions with context
- `restoreSessionState()`: Recreate sessions from serialized state
- Multi-tenant support for SaaS deployments
- Automatic session expiration handling
### Security
- **Important**: API keys exported as plaintext - downstream MUST encrypt
- Session validation on restore prevents invalid state injection
- Respects `sessionTimeout` configuration during restore
## [2.24.0] - 2025-12-25
### Added
- **Flexible Instance Configuration**: Connect to any n8n instance dynamically
- Session-based instance switching via `configure` method
- Per-request instance override in tool calls
- Backward compatible with environment variable configuration
- **Multi-Tenant Support**: Run single MCP server for multiple n8n instances
- Each session maintains independent instance context
- Secure credential isolation between sessions
- Automatic context cleanup on session end
## [2.23.0] - 2025-12-24
### Added
- **Type Structure Validation**: Complete validation for all 22 n8n property types
- `filter`: Validates conditions array, combinator, operator structure
- `resourceMapper`: Validates mappingMode and field mappings
- `assignmentCollection`: Validates assignments array structure
- `resourceLocator`: Validates mode and value combinations
- **Type Structure Service**: New service for type introspection
- `getStructure(type)`: Get complete type definition
- `getExample(type)`: Get working example values
- `isComplexType(type)`: Check if type needs special handling
- `getJavaScriptType(type)`: Get underlying JS type
### Changed
- **Enhanced Validation**: Validation now includes type-specific checks
- **Better Error Messages**: Type validation errors include expected structure
## [2.22.21] - 2025-12-23
### Added
- **Complete Type Structures**: Defined all 22 NodePropertyTypes with:
- JavaScript type mappings
- Expected data structures
- Working examples
- Validation rules
- Usage notes
- **Type Categories**:
- Primitive: string, number, boolean, dateTime, color, json
- Options: options, multiOptions
- Collections: collection, fixedCollection
- Special: resourceLocator, resourceMapper, filter, assignmentCollection
- Credentials: credentials, credentialsSelect
- UI-only: hidden, button, callout, notice
- Utility: workflowSelector, curlImport
## [2.22.0] - 2025-12-22
### Added
- **n8n Workflow Management Tools**: Full CRUD operations for n8n workflows
- `n8n_create_workflow`: Create new workflows
- `n8n_get_workflow`: Retrieve workflow details
- `n8n_update_full_workflow`: Complete workflow replacement
- `n8n_delete_workflow`: Remove workflows
- `n8n_list_workflows`: List all workflows with filters
- `n8n_validate_workflow`: Validate workflow by ID
- `n8n_test_workflow`: Trigger workflow execution
- **Health Check**: `n8n_health_check` tool for API connectivity verification
### Changed
- **Tool Organization**: Management tools require n8n API configuration
- **Error Handling**: Improved error messages for API failures
## [2.21.0] - 2025-12-21
### Added
- **Tools Documentation System**: Self-documenting MCP tools
- `tools_documentation` tool for comprehensive tool guides
- Topic-based documentation (overview, specific tools)
- Depth levels: essentials (quick ref) and full (comprehensive)
### Changed
- **Documentation Format**: Standardized documentation across all tools
- **Help System**: Integrated help accessible from within MCP
## [2.20.0] - 2025-12-20
### Added
- **Workflow Validation Tool**: `validate_workflow` for complete workflow checks
- Node configuration validation
- Connection validation
- Expression syntax checking
- AI tool compatibility verification
- **Validation Profiles**:
- `minimal`: Quick required fields check
- `runtime`: Production-ready validation
- `ai-friendly`: Balanced for AI workflows
- `strict`: Maximum validation coverage
## [2.19.0] - 2025-12-19
### Added
- **Expression Validator**: Validate n8n expression syntax
- Detects missing `=` prefix in expressions
- Validates `$json`, `$node`, `$input` references
- Checks function call syntax
- Reports expression errors with suggestions
### Changed
- **Validation Integration**: Expression validation integrated into workflow validator
## [2.18.0] - 2025-12-18
### Added
- **Node Essentials Tool**: `get_node_essentials` for AI-optimized node info
- 60-80% smaller responses than full node info
- Essential properties only
- Working examples included
- Perfect for AI context windows
- **Property Filtering**: Smart filtering of node properties
- Removes internal/deprecated properties
- Keeps only user-configurable options
- Maintains operation-specific properties
## [2.17.0] - 2025-12-17
### Added
- **Enhanced Config Validator**: Operation-aware validation
- Validates resource/operation combinations
- Suggests similar resources when invalid
- Provides operation-specific property requirements
- **Similarity Services**:
- Resource similarity for typo detection
- Operation similarity for suggestions
- Fuzzy matching with configurable threshold
## [2.16.0] - 2025-12-16
### Added
- **Template System**: Workflow templates from n8n.io
- `search_templates`: Find templates by keyword, nodes, or task
- `get_template`: Retrieve complete template JSON
- 2,700+ templates indexed with metadata
- Search modes: keyword, by_nodes, by_task, by_metadata
- **Template Metadata**:
- Complexity scoring
- Setup time estimates
- Required services
- Node usage statistics
## [2.15.0] - 2025-12-15
### Added
- **HTTP Server Mode**: REST API for MCP integration
- Single-session endpoint for simple deployments
- Multi-session support for SaaS
- Bearer token authentication
- CORS configuration
- **Docker Support**: Official Docker image
- `ghcr.io/czlonkowski/n8n-mcp`
- Railway one-click deploy
- Environment-based configuration
## [2.14.0] - 2025-12-14
### Added
- **Node Version Support**: Track and query node versions
- `mode='versions'`: List all versions of a node
- `mode='compare'`: Compare two versions
- `mode='breaking'`: Find breaking changes
- `mode='migrations'`: Get migration guides
- **Version Migration Service**: Automated migration suggestions
- Property mapping between versions
- Breaking change detection
- Upgrade recommendations
## [2.13.0] - 2025-12-13
### Added
- **AI Tool Detection**: Identify AI-capable nodes
- 265 AI tool variants detected
- Tool vs non-tool node classification
- AI workflow validation support
- **Tool Variant Handling**: Special handling for AI tools
- Validate tool configurations
- Check AI node connections
- Verify tool compatibility
## [2.12.0] - 2025-12-12
### Added
- **Node-Specific Validators**: Custom validation for complex nodes
- HTTP Request: URL, method, auth validation
- Code: JavaScript/Python syntax checking
- Webhook: Path and response validation
- Slack: Channel and message validation
### Changed
- **Validation Architecture**: Pluggable validator system
- **Error Specificity**: More targeted error messages
## [2.11.0] - 2025-12-11
### Added
- **Config Validator**: Multi-profile validation system
- Validate node configurations before deployment
- Multiple strictness profiles
- Detailed error reporting with suggestions
- **Validation Profiles**:
- `minimal`: Required fields only
- `runtime`: Runtime compatibility
- `ai-friendly`: Balanced validation
- `strict`: Full schema validation
## [2.10.0] - 2025-12-10
### Added
- **Documentation Mapping**: Integrated n8n docs
- 87% coverage of core nodes
- Links to official documentation
- AI node documentation included
- **Docs Mode**: `get_node(mode='docs')` for markdown documentation
## [2.9.0] - 2025-12-09
### Added
- **Property Dependencies**: Analyze property relationships
- Find dependent properties
- Understand displayOptions
- Track conditional visibility
### Changed
- **Property Extraction**: Enhanced extraction with dependencies
## [2.8.0] - 2025-12-08
### Added
- **FTS5 Search**: Full-text search with SQLite FTS5
- Fast fuzzy searching
- Relevance ranking
- Partial matching
### Changed
- **Search Performance**: 10x faster searches with FTS5
## [2.7.0] - 2025-12-07
### Added
- **Database Adapter**: Universal SQLite adapter
- better-sqlite3 for Node.js
- sql.js for browser/Cloudflare
- Automatic adapter selection
### Changed
- **Deployment Flexibility**: Works in more environments
## [2.6.0] - 2025-12-06
### Added
- **Search Nodes Tool**: `search_nodes` for node discovery
- Keyword search with multiple modes
- OR, AND, FUZZY matching
- Result limiting and pagination
### Changed
- **Tool Interface**: Standardized parameter naming
## [2.5.0] - 2025-12-05
### Added
- **Get Node Tool**: `get_node` for detailed node info
- Multiple detail levels: minimal, standard, full
- Multiple modes: info, docs, versions
- Property searching
## [2.4.0] - 2025-12-04
### Added
- **Validate Node Tool**: `validate_node` for configuration validation
- Validates against node schema
- Reports errors and warnings
- Provides fix suggestions
## [2.3.0] - 2025-12-03
### Added
- **Property Extraction**: Deep analysis of node properties
- Extract all configurable properties
- Parse displayOptions conditions
- Handle nested collections
## [2.2.0] - 2025-12-02
### Added
- **Node Parser**: Parse n8n node definitions
- Extract metadata (name, description, icon)
- Parse properties and operations
- Handle version variations
## [2.1.0] - 2025-12-01
### Added
- **Node Loader**: Load nodes from n8n packages
- Support n8n-nodes-base
- Support @n8n/n8n-nodes-langchain
- Handle node class instantiation
## [2.0.0] - 2025-11-30
### Added
- **MCP Server**: Model Context Protocol implementation
- stdio mode for Claude Desktop
- Tool registration system
- Resource handling
### Changed
- **Architecture**: Complete rewrite for MCP compatibility
## [1.0.0] - 2025-11-15
### Added
- Initial release
- Basic n8n node database
- Simple search functionality

View File

@@ -14,7 +14,7 @@ RUN --mount=type=cache,target=/root/.npm \
echo '{}' > package.json && \
npm install --no-save typescript@^5.8.3 @types/node@^22.15.30 @types/express@^5.0.3 \
@modelcontextprotocol/sdk@1.20.1 dotenv@^16.5.0 express@^5.1.0 axios@^1.10.0 \
n8n-workflow@^1.96.0 uuid@^11.0.5 @types/uuid@^10.0.0 \
n8n-workflow@^2.4.2 uuid@^11.0.5 @types/uuid@^10.0.0 \
openai@^4.77.0 zod@3.24.1 lru-cache@^11.2.1 @supabase/supabase-js@^2.57.4
# Copy source and build

View File

@@ -5,7 +5,7 @@
[![npm version](https://img.shields.io/npm/v/n8n-mcp.svg)](https://www.npmjs.com/package/n8n-mcp)
[![codecov](https://codecov.io/gh/czlonkowski/n8n-mcp/graph/badge.svg?token=YOUR_TOKEN)](https://codecov.io/gh/czlonkowski/n8n-mcp)
[![Tests](https://img.shields.io/badge/tests-3336%20passing-brightgreen.svg)](https://github.com/czlonkowski/n8n-mcp/actions)
[![n8n version](https://img.shields.io/badge/n8n-2.3.3-orange.svg)](https://github.com/n8n-io/n8n)
[![n8n version](https://img.shields.io/badge/n8n-2.4.4-orange.svg)](https://github.com/n8n-io/n8n)
[![Docker](https://img.shields.io/badge/docker-ghcr.io%2Fczlonkowski%2Fn8n--mcp-green.svg)](https://github.com/czlonkowski/n8n-mcp/pkgs/container/n8n-mcp)
[![Deploy on Railway](https://railway.com/button.svg)](https://railway.com/deploy/n8n-mcp?referralCode=n8n-mcp)

Binary file not shown.

View File

@@ -1 +1 @@
{"version":3,"file":"type-structures.d.ts","sourceRoot":"","sources":["../../src/constants/type-structures.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAe9D,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,iBAAiB,EAAE,aAAa,CAilBpE,CAAC;AAUF,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA4GjC,CAAC"}
{"version":3,"file":"type-structures.d.ts","sourceRoot":"","sources":["../../src/constants/type-structures.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,cAAc,CAAC;AACtD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AAe9D,eAAO,MAAM,eAAe,EAAE,MAAM,CAAC,iBAAiB,EAAE,aAAa,CAkmBpE,CAAC;AAUF,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA4GjC,CAAC"}

View File

@@ -545,6 +545,22 @@ exports.TYPE_STRUCTURES = {
'One-time import feature',
],
},
icon: {
type: 'primitive',
jsType: 'string',
description: 'Icon identifier for visual representation',
example: 'fa:envelope',
examples: ['fa:envelope', 'fa:user', 'fa:cog', 'file:slack.svg'],
validation: {
allowEmpty: false,
allowExpressions: false,
},
notes: [
'References icon by name or file path',
'Supports Font Awesome icons (fa:) and file paths (file:)',
'Used for visual customization in UI',
],
},
};
exports.COMPLEX_TYPE_EXAMPLES = {
collection: {

File diff suppressed because one or more lines are too long

3972
package-lock.json generated

File diff suppressed because it is too large Load Diff

View File

@@ -1,6 +1,6 @@
{
"name": "n8n-mcp",
"version": "2.33.2",
"version": "2.33.5",
"description": "Integration between n8n workflow automation and Model Context Protocol (MCP)",
"main": "dist/index.js",
"types": "dist/index.d.ts",
@@ -150,16 +150,16 @@
},
"dependencies": {
"@modelcontextprotocol/sdk": "1.20.1",
"@n8n/n8n-nodes-langchain": "^2.3.2",
"@n8n/n8n-nodes-langchain": "^2.4.3",
"@supabase/supabase-js": "^2.57.4",
"dotenv": "^16.5.0",
"express": "^5.1.0",
"express-rate-limit": "^7.1.5",
"form-data": "^4.0.5",
"lru-cache": "^11.2.1",
"n8n": "^2.3.3",
"n8n-core": "^2.3.2",
"n8n-workflow": "^2.3.2",
"n8n": "^2.4.4",
"n8n-core": "^2.4.2",
"n8n-workflow": "^2.4.2",
"openai": "^4.77.0",
"sql.js": "^1.13.0",
"tslib": "^2.6.2",

View File

@@ -5,7 +5,7 @@
* These structures define the expected data format, JavaScript type,
* validation rules, and examples for each property type.
*
* Based on n8n-workflow v1.120.3 NodePropertyTypes
* Based on n8n-workflow v2.4.2 NodePropertyTypes
*
* @module constants/type-structures
* @since 2.23.0
@@ -15,7 +15,7 @@ import type { NodePropertyTypes } from 'n8n-workflow';
import type { TypeStructure } from '../types/type-structures';
/**
* Complete type structure definitions for all 22 NodePropertyTypes
* Complete type structure definitions for all 23 NodePropertyTypes
*
* Each entry defines:
* - type: Category (primitive/object/collection/special)
@@ -620,6 +620,23 @@ export const TYPE_STRUCTURES: Record<NodePropertyTypes, TypeStructure> = {
'One-time import feature',
],
},
icon: {
type: 'primitive',
jsType: 'string',
description: 'Icon identifier for visual representation',
example: 'fa:envelope',
examples: ['fa:envelope', 'fa:user', 'fa:cog', 'file:slack.svg'],
validation: {
allowEmpty: false,
allowExpressions: false,
},
notes: [
'References icon by name or file path',
'Supports Font Awesome icons (fa:) and file paths (file:)',
'Used for visual customization in UI',
],
},
};
/**

View File

@@ -419,12 +419,36 @@ class BetterSQLiteStatement implements PreparedStatement {
/**
* Statement wrapper for sql.js
*
* IMPORTANT: sql.js requires explicit memory management via Statement.free().
* This wrapper automatically frees statement memory after each operation
* to prevent memory leaks during sustained traffic.
*
* See: https://sql.js.org/documentation/Statement.html
* "After calling db.prepare() you must manually free the assigned memory
* by calling Statement.free()."
*/
class SQLJSStatement implements PreparedStatement {
private boundParams: any = null;
private freed: boolean = false;
constructor(private stmt: any, private onModify: () => void) {}
/**
* Free the underlying sql.js statement memory.
* Safe to call multiple times - subsequent calls are no-ops.
*/
private freeStatement(): void {
if (!this.freed && this.stmt) {
try {
this.stmt.free();
this.freed = true;
} catch (e) {
// Statement may already be freed or invalid - ignore
}
}
}
run(...params: any[]): RunResult {
try {
if (params.length > 0) {
@@ -433,10 +457,10 @@ class SQLJSStatement implements PreparedStatement {
this.stmt.bind(this.boundParams);
}
}
this.stmt.run();
this.onModify();
// sql.js doesn't provide changes/lastInsertRowid easily
return {
changes: 1, // Assume success means 1 change
@@ -445,9 +469,12 @@ class SQLJSStatement implements PreparedStatement {
} catch (error) {
this.stmt.reset();
throw error;
} finally {
// Free statement memory after write operation completes
this.freeStatement();
}
}
get(...params: any[]): any {
try {
if (params.length > 0) {
@@ -456,21 +483,24 @@ class SQLJSStatement implements PreparedStatement {
this.stmt.bind(this.boundParams);
}
}
if (this.stmt.step()) {
const result = this.stmt.getAsObject();
this.stmt.reset();
return this.convertIntegerColumns(result);
}
this.stmt.reset();
return undefined;
} catch (error) {
this.stmt.reset();
throw error;
} finally {
// Free statement memory after read operation completes
this.freeStatement();
}
}
all(...params: any[]): any[] {
try {
if (params.length > 0) {
@@ -479,17 +509,20 @@ class SQLJSStatement implements PreparedStatement {
this.stmt.bind(this.boundParams);
}
}
const results: any[] = [];
while (this.stmt.step()) {
results.push(this.convertIntegerColumns(this.stmt.getAsObject()));
}
this.stmt.reset();
return results;
} catch (error) {
this.stmt.reset();
throw error;
} finally {
// Free statement memory after read operation completes
this.freeStatement();
}
}

View File

@@ -0,0 +1,197 @@
/**
* Shared Database Manager - Singleton for cross-session database connection
*
* This module implements a singleton pattern to share a single database connection
* across all MCP server sessions. This prevents memory leaks caused by each session
* creating its own database connection (~900MB per session).
*
* Memory impact: Reduces per-session memory from ~900MB to near-zero by sharing
* a single ~68MB database connection across all sessions.
*
* Issue: https://github.com/czlonkowski/n8n-mcp/issues/XXX
*/
import { DatabaseAdapter, createDatabaseAdapter } from './database-adapter';
import { NodeRepository } from './node-repository';
import { TemplateService } from '../templates/template-service';
import { EnhancedConfigValidator } from '../services/enhanced-config-validator';
import { logger } from '../utils/logger';
/**
* Shared database state - holds the singleton connection and services
*/
export interface SharedDatabaseState {
db: DatabaseAdapter;
repository: NodeRepository;
templateService: TemplateService;
dbPath: string;
refCount: number;
initialized: boolean;
}
// Module-level singleton state
let sharedState: SharedDatabaseState | null = null;
let initializationPromise: Promise<SharedDatabaseState> | null = null;
/**
* Get or create the shared database connection
*
* Thread-safe initialization using a promise lock pattern.
* Multiple concurrent calls will wait for the same initialization.
*
* @param dbPath - Path to the SQLite database file
* @returns Shared database state with connection and services
*/
export async function getSharedDatabase(dbPath: string): Promise<SharedDatabaseState> {
// If already initialized with the same path, increment ref count and return
if (sharedState && sharedState.initialized && sharedState.dbPath === dbPath) {
sharedState.refCount++;
logger.debug('Reusing shared database connection', {
refCount: sharedState.refCount,
dbPath
});
return sharedState;
}
// If already initialized with a DIFFERENT path, this is a configuration error
if (sharedState && sharedState.initialized && sharedState.dbPath !== dbPath) {
logger.error('Attempted to initialize shared database with different path', {
existingPath: sharedState.dbPath,
requestedPath: dbPath
});
throw new Error(`Shared database already initialized with different path: ${sharedState.dbPath}`);
}
// If initialization is in progress, wait for it
if (initializationPromise) {
try {
const state = await initializationPromise;
state.refCount++;
logger.debug('Reusing shared database (waited for init)', {
refCount: state.refCount,
dbPath
});
return state;
} catch (error) {
// Initialization failed while we were waiting, clear promise and rethrow
initializationPromise = null;
throw error;
}
}
// Start new initialization
initializationPromise = initializeSharedDatabase(dbPath);
try {
const state = await initializationPromise;
// Clear the promise on success to allow future re-initialization after close
initializationPromise = null;
return state;
} catch (error) {
// Clear promise on failure to allow retry
initializationPromise = null;
throw error;
}
}
/**
* Initialize the shared database connection and services
*/
async function initializeSharedDatabase(dbPath: string): Promise<SharedDatabaseState> {
logger.info('Initializing shared database connection', { dbPath });
const db = await createDatabaseAdapter(dbPath);
const repository = new NodeRepository(db);
const templateService = new TemplateService(db);
// Initialize similarity services for enhanced validation
EnhancedConfigValidator.initializeSimilarityServices(repository);
sharedState = {
db,
repository,
templateService,
dbPath,
refCount: 1,
initialized: true
};
logger.info('Shared database initialized successfully', {
dbPath,
refCount: sharedState.refCount
});
return sharedState;
}
/**
* Release a reference to the shared database
*
* Decrements the reference count. Does NOT close the database
* as it's shared across all sessions for the lifetime of the process.
*
* @param state - The shared database state to release
*/
export function releaseSharedDatabase(state: SharedDatabaseState): void {
if (!state || !sharedState) {
return;
}
// Guard against double-release (refCount going negative)
if (sharedState.refCount <= 0) {
logger.warn('Attempted to release shared database with refCount already at or below 0', {
refCount: sharedState.refCount
});
return;
}
sharedState.refCount--;
logger.debug('Released shared database reference', {
refCount: sharedState.refCount
});
// Note: We intentionally do NOT close the database even when refCount hits 0
// The database should remain open for the lifetime of the process to handle
// new sessions. Only process shutdown should close it.
}
/**
* Force close the shared database (for graceful shutdown only)
*
* This should only be called during process shutdown, not during normal
* session cleanup. Closing the database would break other active sessions.
*/
export async function closeSharedDatabase(): Promise<void> {
if (!sharedState) {
return;
}
logger.info('Closing shared database connection', {
refCount: sharedState.refCount
});
try {
sharedState.db.close();
} catch (error) {
logger.warn('Error closing shared database', {
error: error instanceof Error ? error.message : String(error)
});
}
sharedState = null;
initializationPromise = null;
}
/**
* Check if shared database is initialized
*/
export function isSharedDatabaseInitialized(): boolean {
return sharedState !== null && sharedState.initialized;
}
/**
* Get current reference count (for debugging/monitoring)
*/
export function getSharedDatabaseRefCount(): number {
return sharedState?.refCount ?? 0;
}

View File

@@ -26,6 +26,7 @@ import {
} from './utils/protocol-version';
import { InstanceContext, validateInstanceContext } from './types/instance-context';
import { SessionState } from './types/session-state';
import { closeSharedDatabase } from './database/shared-database';
dotenv.config();
@@ -106,7 +107,12 @@ export class SingleSessionHTTPServer {
private session: Session | null = null; // Keep for SSE compatibility
private consoleManager = new ConsoleManager();
private expressServer: any;
private sessionTimeout = 30 * 60 * 1000; // 30 minutes
// Session timeout reduced from 30 minutes to 5 minutes for faster cleanup
// Configurable via SESSION_TIMEOUT_MINUTES environment variable
// This prevents memory buildup from stale sessions
private sessionTimeout = parseInt(
process.env.SESSION_TIMEOUT_MINUTES || '5', 10
) * 60 * 1000;
private authToken: string | null = null;
private cleanupTimer: NodeJS.Timeout | null = null;
@@ -492,6 +498,29 @@ export class SingleSessionHTTPServer {
// For initialize requests: always create new transport and server
logger.info('handleRequest: Creating new transport for initialize request');
// EAGER CLEANUP: Remove existing sessions for the same instance
// This prevents memory buildup when clients reconnect without proper cleanup
if (instanceContext?.instanceId) {
const sessionsToRemove: string[] = [];
for (const [existingSessionId, context] of Object.entries(this.sessionContexts)) {
if (context?.instanceId === instanceContext.instanceId) {
sessionsToRemove.push(existingSessionId);
}
}
for (const oldSessionId of sessionsToRemove) {
// Double-check session still exists (may have been cleaned by concurrent request)
if (!this.transports[oldSessionId]) {
continue;
}
logger.info('Cleaning up previous session for instance', {
instanceId: instanceContext.instanceId,
oldSession: oldSessionId,
reason: 'instance_reconnect'
});
await this.removeSession(oldSessionId, 'instance_reconnect');
}
}
// Generate session ID based on multi-tenant configuration
let sessionIdToUse: string;
@@ -677,11 +706,25 @@ export class SingleSessionHTTPServer {
private async resetSessionSSE(res: express.Response): Promise<void> {
// Clean up old session if exists
if (this.session) {
const sessionId = this.session.sessionId;
logger.info('Closing previous session for SSE', { sessionId });
// Close server first to free resources (database, cache timer, etc.)
// This mirrors the cleanup pattern in removeSession() (issue #542)
// Handle server close errors separately so transport close still runs
if (this.session.server && typeof this.session.server.close === 'function') {
try {
await this.session.server.close();
} catch (serverError) {
logger.warn('Error closing server for SSE session', { sessionId, error: serverError });
}
}
// Close transport last - always attempt even if server.close() failed
try {
logger.info('Closing previous session for SSE', { sessionId: this.session.sessionId });
await this.session.transport.close();
} catch (error) {
logger.warn('Error closing previous session:', error);
} catch (transportError) {
logger.warn('Error closing transport for SSE session', { sessionId, error: transportError });
}
}
@@ -1408,7 +1451,16 @@ export class SingleSessionHTTPServer {
});
});
}
// Close the shared database connection (only during process shutdown)
// This must happen after all sessions are closed
try {
await closeSharedDatabase();
logger.info('Shared database closed');
} catch (error) {
logger.warn('Error closing shared database:', error);
}
logger.info('Single-Session HTTP server shutdown completed');
}

View File

@@ -14,6 +14,7 @@ import { getWorkflowExampleString } from './workflow-examples';
import { logger } from '../utils/logger';
import { NodeRepository } from '../database/node-repository';
import { DatabaseAdapter, createDatabaseAdapter } from '../database/database-adapter';
import { getSharedDatabase, releaseSharedDatabase, SharedDatabaseState } from '../database/shared-database';
import { PropertyFilter } from '../services/property-filter';
import { TaskTemplates } from '../services/task-templates';
import { ConfigValidator } from '../services/config-validator';
@@ -150,6 +151,9 @@ export class N8NDocumentationMCPServer {
private previousToolTimestamp: number = Date.now();
private earlyLogger: EarlyErrorLogger | null = null;
private disabledToolsCache: Set<string> | null = null;
private useSharedDatabase: boolean = false; // Track if using shared DB for cleanup
private sharedDbState: SharedDatabaseState | null = null; // Reference to shared DB state for release
private isShutdown: boolean = false; // Prevent double-shutdown
constructor(instanceContext?: InstanceContext, earlyLogger?: EarlyErrorLogger) {
this.instanceContext = instanceContext;
@@ -245,18 +249,39 @@ export class N8NDocumentationMCPServer {
* Order of cleanup:
* 1. Close MCP server connection
* 2. Destroy cache (clears entries AND stops cleanup timer)
* 3. Close database connection
* 3. Release shared database OR close dedicated connection
* 4. Null out references to help GC
*
* IMPORTANT: For shared databases, we only release the reference (decrement refCount),
* NOT close the database. The database stays open for other sessions.
* For in-memory databases (tests), we close the dedicated connection.
*/
async close(): Promise<void> {
// Wait for initialization to complete (or fail) before cleanup
// This prevents race conditions where close runs while init is in progress
try {
await this.initialized;
} catch (error) {
// Initialization failed - that's OK, we still need to clean up
logger.debug('Initialization had failed, proceeding with cleanup', {
error: error instanceof Error ? error.message : String(error)
});
}
try {
await this.server.close();
// Use destroy() not clear() - also stops the cleanup timer
this.cache.destroy();
// Close database connection before nullifying reference
if (this.db) {
// Handle database cleanup based on whether it's shared or dedicated
if (this.useSharedDatabase && this.sharedDbState) {
// Shared database: release reference, don't close
// The database stays open for other sessions
releaseSharedDatabase(this.sharedDbState);
logger.debug('Released shared database reference');
} else if (this.db) {
// Dedicated database (in-memory for tests): close it
try {
this.db.close();
} catch (dbError) {
@@ -271,6 +296,7 @@ export class N8NDocumentationMCPServer {
this.repository = null;
this.templateService = null;
this.earlyLogger = null;
this.sharedDbState = null;
} catch (error) {
// Log but don't throw - cleanup should be best-effort
logger.warn('Error closing MCP server', { error: error instanceof Error ? error.message : String(error) });
@@ -286,23 +312,32 @@ export class N8NDocumentationMCPServer {
logger.debug('Database initialization starting...', { dbPath });
this.db = await createDatabaseAdapter(dbPath);
logger.debug('Database adapter created');
// If using in-memory database for tests, initialize schema
// For in-memory databases (tests), create a dedicated connection
// For regular databases, use the shared connection to prevent memory leaks
if (dbPath === ':memory:') {
this.db = await createDatabaseAdapter(dbPath);
logger.debug('Database adapter created (in-memory mode)');
await this.initializeInMemorySchema();
logger.debug('In-memory schema initialized');
this.repository = new NodeRepository(this.db);
this.templateService = new TemplateService(this.db);
// Initialize similarity services for enhanced validation
EnhancedConfigValidator.initializeSimilarityServices(this.repository);
this.useSharedDatabase = false;
} else {
// Use shared database connection to prevent ~900MB memory leak per session
// See: Memory leak fix - database was being duplicated per session
const sharedState = await getSharedDatabase(dbPath);
this.db = sharedState.db;
this.repository = sharedState.repository;
this.templateService = sharedState.templateService;
this.sharedDbState = sharedState;
this.useSharedDatabase = true;
logger.debug('Using shared database connection');
}
this.repository = new NodeRepository(this.db);
logger.debug('Node repository initialized');
this.templateService = new TemplateService(this.db);
logger.debug('Template service initialized');
// Initialize similarity services for enhanced validation
EnhancedConfigValidator.initializeSimilarityServices(this.repository);
logger.debug('Similarity services initialized');
// Checkpoint: Database connected (v2.18.3)
@@ -3910,8 +3945,33 @@ Full documentation is being prepared. For now, use get_node_essentials for confi
}
async shutdown(): Promise<void> {
// Prevent double-shutdown
if (this.isShutdown) {
logger.debug('Shutdown already called, skipping');
return;
}
this.isShutdown = true;
logger.info('Shutting down MCP server...');
// Wait for initialization to complete (or fail) before cleanup
// This prevents race conditions where shutdown runs while init is in progress
try {
await this.initialized;
} catch (error) {
// Initialization failed - that's OK, we still need to clean up
logger.debug('Initialization had failed, proceeding with cleanup', {
error: error instanceof Error ? error.message : String(error)
});
}
// Close MCP server connection (for consistency with close() method)
try {
await this.server.close();
} catch (error) {
logger.error('Error closing MCP server:', error);
}
// Clean up cache timers to prevent memory leaks
if (this.cache) {
try {
@@ -3921,15 +3981,31 @@ Full documentation is being prepared. For now, use get_node_essentials for confi
logger.error('Error cleaning up cache:', error);
}
}
// Close database connection if it exists
if (this.db) {
// Handle database cleanup based on whether it's shared or dedicated
// For shared databases, we only release the reference (decrement refCount)
// For dedicated databases (in-memory for tests), we close the connection
if (this.useSharedDatabase && this.sharedDbState) {
try {
await this.db.close();
releaseSharedDatabase(this.sharedDbState);
logger.info('Released shared database reference');
} catch (error) {
logger.error('Error releasing shared database:', error);
}
} else if (this.db) {
try {
this.db.close();
logger.info('Database connection closed');
} catch (error) {
logger.error('Error closing database:', error);
}
}
// Null out references to help garbage collection
this.db = null;
this.repository = null;
this.templateService = null;
this.earlyLogger = null;
this.sharedDbState = null;
}
}

View File

@@ -90,7 +90,7 @@ export class TypeStructureService {
/**
* Get all type structure definitions
*
* Returns a record of all 22 NodePropertyTypes with their structures.
* Returns a record of all 23 NodePropertyTypes with their structures.
* Useful for documentation, validation setup, or UI generation.
*
* @returns Record mapping all types to their structures

View File

@@ -58,6 +58,13 @@ export class TelemetryBatchProcessor {
private flushTimes: number[] = [];
private deadLetterQueue: (TelemetryEvent | WorkflowTelemetry | WorkflowMutationRecord)[] = [];
private readonly maxDeadLetterSize = 100;
// Track event listeners for proper cleanup to prevent memory leaks
private eventListeners: {
beforeExit?: () => void;
sigint?: () => void;
sigterm?: () => void;
} = {};
private started: boolean = false;
constructor(
private supabase: SupabaseClient | null,
@@ -72,6 +79,12 @@ export class TelemetryBatchProcessor {
start(): void {
if (!this.isEnabled() || !this.supabase) return;
// Guard against multiple starts (prevents event listener accumulation)
if (this.started) {
logger.debug('Telemetry batch processor already started, skipping');
return;
}
// Set up periodic flushing
this.flushTimer = setInterval(() => {
this.flush();
@@ -83,17 +96,22 @@ export class TelemetryBatchProcessor {
this.flushTimer.unref();
}
// Set up process exit handlers
process.on('beforeExit', () => this.flush());
process.on('SIGINT', () => {
// Set up process exit handlers with stored references for cleanup
this.eventListeners.beforeExit = () => this.flush();
this.eventListeners.sigint = () => {
this.flush();
process.exit(0);
});
process.on('SIGTERM', () => {
};
this.eventListeners.sigterm = () => {
this.flush();
process.exit(0);
});
};
process.on('beforeExit', this.eventListeners.beforeExit);
process.on('SIGINT', this.eventListeners.sigint);
process.on('SIGTERM', this.eventListeners.sigterm);
this.started = true;
logger.debug('Telemetry batch processor started');
}
@@ -105,6 +123,20 @@ export class TelemetryBatchProcessor {
clearInterval(this.flushTimer);
this.flushTimer = undefined;
}
// Remove event listeners to prevent memory leaks
if (this.eventListeners.beforeExit) {
process.removeListener('beforeExit', this.eventListeners.beforeExit);
}
if (this.eventListeners.sigint) {
process.removeListener('SIGINT', this.eventListeners.sigint);
}
if (this.eventListeners.sigterm) {
process.removeListener('SIGTERM', this.eventListeners.sigterm);
}
this.eventListeners = {};
this.started = false;
logger.debug('Telemetry batch processor stopped');
}

View File

@@ -11,7 +11,7 @@ import { isTypeStructure } from '@/types/type-structures';
import type { NodePropertyTypes } from 'n8n-workflow';
describe('TYPE_STRUCTURES', () => {
// All 22 NodePropertyTypes from n8n-workflow
// All 23 NodePropertyTypes from n8n-workflow
const ALL_PROPERTY_TYPES: NodePropertyTypes[] = [
'boolean',
'button',
@@ -20,6 +20,7 @@ describe('TYPE_STRUCTURES', () => {
'dateTime',
'fixedCollection',
'hidden',
'icon',
'json',
'callout',
'notice',
@@ -38,16 +39,16 @@ describe('TYPE_STRUCTURES', () => {
];
describe('Completeness', () => {
it('should define all 22 NodePropertyTypes', () => {
it('should define all 23 NodePropertyTypes', () => {
const definedTypes = Object.keys(TYPE_STRUCTURES);
expect(definedTypes).toHaveLength(22);
expect(definedTypes).toHaveLength(23);
for (const type of ALL_PROPERTY_TYPES) {
expect(TYPE_STRUCTURES).toHaveProperty(type);
}
});
it('should not have extra types beyond the 22 standard types', () => {
it('should not have extra types beyond the 23 standard types', () => {
const definedTypes = Object.keys(TYPE_STRUCTURES);
const extraTypes = definedTypes.filter((type) => !ALL_PROPERTY_TYPES.includes(type as NodePropertyTypes));

View File

@@ -0,0 +1,302 @@
import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
// Mock dependencies at module level
const mockDb = {
prepare: vi.fn().mockReturnValue({
get: vi.fn(),
all: vi.fn(),
run: vi.fn()
}),
exec: vi.fn(),
close: vi.fn(),
pragma: vi.fn(),
inTransaction: false,
transaction: vi.fn(),
checkFTS5Support: vi.fn()
};
vi.mock('../../../src/database/database-adapter', () => ({
createDatabaseAdapter: vi.fn().mockResolvedValue(mockDb)
}));
vi.mock('../../../src/database/node-repository', () => ({
NodeRepository: vi.fn().mockImplementation(() => ({
getNodeTypes: vi.fn().mockReturnValue([])
}))
}));
vi.mock('../../../src/templates/template-service', () => ({
TemplateService: vi.fn().mockImplementation(() => ({}))
}));
vi.mock('../../../src/services/enhanced-config-validator', () => ({
EnhancedConfigValidator: {
initializeSimilarityServices: vi.fn()
}
}));
vi.mock('../../../src/utils/logger', () => ({
logger: {
debug: vi.fn(),
info: vi.fn(),
warn: vi.fn(),
error: vi.fn()
}
}));
describe('Shared Database Module', () => {
let sharedDbModule: typeof import('../../../src/database/shared-database');
let createDatabaseAdapter: ReturnType<typeof vi.fn>;
beforeEach(async () => {
// Reset all mocks
vi.clearAllMocks();
mockDb.close.mockReset();
// Reset modules to get fresh state
vi.resetModules();
// Import fresh module
sharedDbModule = await import('../../../src/database/shared-database');
// Get the mocked function
const adapterModule = await import('../../../src/database/database-adapter');
createDatabaseAdapter = adapterModule.createDatabaseAdapter as ReturnType<typeof vi.fn>;
});
afterEach(async () => {
// Clean up any shared state by closing
try {
await sharedDbModule.closeSharedDatabase();
} catch {
// Ignore errors during cleanup
}
});
describe('getSharedDatabase', () => {
it('should initialize database on first call', async () => {
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state).toBeDefined();
expect(state.db).toBe(mockDb);
expect(state.dbPath).toBe('/path/to/db');
expect(state.refCount).toBe(1);
expect(state.initialized).toBe(true);
expect(createDatabaseAdapter).toHaveBeenCalledWith('/path/to/db');
});
it('should reuse existing connection and increment refCount', async () => {
// First call initializes
const state1 = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state1.refCount).toBe(1);
// Second call reuses
const state2 = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state2.refCount).toBe(2);
// Same object
expect(state1).toBe(state2);
// Only initialized once
expect(createDatabaseAdapter).toHaveBeenCalledTimes(1);
});
it('should throw error when called with different path', async () => {
await sharedDbModule.getSharedDatabase('/path/to/db1');
await expect(sharedDbModule.getSharedDatabase('/path/to/db2'))
.rejects.toThrow('Shared database already initialized with different path');
});
it('should handle concurrent initialization requests', async () => {
// Start two requests concurrently
const [state1, state2] = await Promise.all([
sharedDbModule.getSharedDatabase('/path/to/db'),
sharedDbModule.getSharedDatabase('/path/to/db')
]);
// Both should get the same state
expect(state1).toBe(state2);
// RefCount should be 2 (one for each call)
expect(state1.refCount).toBe(2);
// Only one actual initialization
expect(createDatabaseAdapter).toHaveBeenCalledTimes(1);
});
it('should handle initialization failure', async () => {
createDatabaseAdapter.mockRejectedValueOnce(new Error('DB error'));
await expect(sharedDbModule.getSharedDatabase('/path/to/db'))
.rejects.toThrow('DB error');
// After failure, should not be initialized
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(false);
});
it('should allow retry after initialization failure', async () => {
// First call fails
createDatabaseAdapter.mockRejectedValueOnce(new Error('DB error'));
await expect(sharedDbModule.getSharedDatabase('/path/to/db'))
.rejects.toThrow('DB error');
// Reset mock for successful call
createDatabaseAdapter.mockResolvedValueOnce(mockDb);
// Second call succeeds
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state).toBeDefined();
expect(state.initialized).toBe(true);
});
});
describe('releaseSharedDatabase', () => {
it('should decrement refCount', async () => {
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state.refCount).toBe(1);
sharedDbModule.releaseSharedDatabase(state);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
});
it('should not decrement below 0', async () => {
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
// Release once (refCount: 1 -> 0)
sharedDbModule.releaseSharedDatabase(state);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
// Release again (should stay at 0, not go negative)
sharedDbModule.releaseSharedDatabase(state);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
});
it('should handle null state gracefully', () => {
// Should not throw
sharedDbModule.releaseSharedDatabase(null as any);
});
it('should not close database when refCount hits 0', async () => {
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
sharedDbModule.releaseSharedDatabase(state);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
expect(mockDb.close).not.toHaveBeenCalled();
// Database should still be accessible
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(true);
});
});
describe('closeSharedDatabase', () => {
it('should close database and clear state', async () => {
// Get state
await sharedDbModule.getSharedDatabase('/path/to/db');
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(true);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(1);
await sharedDbModule.closeSharedDatabase();
// State should be cleared
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(false);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
});
it('should handle close error gracefully', async () => {
await sharedDbModule.getSharedDatabase('/path/to/db');
mockDb.close.mockImplementationOnce(() => {
throw new Error('Close error');
});
// Should not throw
await sharedDbModule.closeSharedDatabase();
// State should still be cleared
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(false);
});
it('should be idempotent when already closed', async () => {
// Close without ever initializing
await sharedDbModule.closeSharedDatabase();
// Should not throw
await sharedDbModule.closeSharedDatabase();
});
it('should allow re-initialization after close', async () => {
// Initialize
const state1 = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state1.refCount).toBe(1);
// Close
await sharedDbModule.closeSharedDatabase();
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(false);
// Re-initialize
const state2 = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state2.refCount).toBe(1);
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(true);
// Should be a new state object
expect(state1).not.toBe(state2);
});
});
describe('isSharedDatabaseInitialized', () => {
it('should return false before initialization', () => {
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(false);
});
it('should return true after initialization', async () => {
await sharedDbModule.getSharedDatabase('/path/to/db');
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(true);
});
it('should return false after close', async () => {
await sharedDbModule.getSharedDatabase('/path/to/db');
await sharedDbModule.closeSharedDatabase();
expect(sharedDbModule.isSharedDatabaseInitialized()).toBe(false);
});
});
describe('getSharedDatabaseRefCount', () => {
it('should return 0 before initialization', () => {
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
});
it('should return correct refCount after multiple operations', async () => {
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(1);
await sharedDbModule.getSharedDatabase('/path/to/db');
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(2);
await sharedDbModule.getSharedDatabase('/path/to/db');
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(3);
sharedDbModule.releaseSharedDatabase(state);
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(2);
});
it('should return 0 after close', async () => {
await sharedDbModule.getSharedDatabase('/path/to/db');
await sharedDbModule.closeSharedDatabase();
expect(sharedDbModule.getSharedDatabaseRefCount()).toBe(0);
});
});
describe('SharedDatabaseState interface', () => {
it('should expose correct properties', async () => {
const state = await sharedDbModule.getSharedDatabase('/path/to/db');
expect(state).toHaveProperty('db');
expect(state).toHaveProperty('repository');
expect(state).toHaveProperty('templateService');
expect(state).toHaveProperty('dbPath');
expect(state).toHaveProperty('refCount');
expect(state).toHaveProperty('initialized');
});
});
});

View File

@@ -333,13 +333,14 @@ describe('HTTP Server Session Management', () => {
server = new SingleSessionHTTPServer();
// Mock expired sessions
// Note: Default session timeout is 5 minutes (configurable via SESSION_TIMEOUT_MINUTES)
const mockSessionMetadata = {
'session-1': {
lastAccess: new Date(Date.now() - 40 * 60 * 1000), // 40 minutes ago (expired)
'session-1': {
lastAccess: new Date(Date.now() - 10 * 60 * 1000), // 10 minutes ago (expired with 5 min timeout)
createdAt: new Date(Date.now() - 60 * 60 * 1000)
},
'session-2': {
lastAccess: new Date(Date.now() - 10 * 60 * 1000), // 10 minutes ago (not expired)
'session-2': {
lastAccess: new Date(Date.now() - 2 * 60 * 1000), // 2 minutes ago (not expired with 5 min timeout)
createdAt: new Date(Date.now() - 20 * 60 * 1000)
}
};
@@ -514,15 +515,16 @@ describe('HTTP Server Session Management', () => {
it('should get session metrics correctly', async () => {
server = new SingleSessionHTTPServer();
// Note: Default session timeout is 5 minutes (configurable via SESSION_TIMEOUT_MINUTES)
const now = Date.now();
(server as any).sessionMetadata = {
'active-session': {
lastAccess: new Date(now - 10 * 60 * 1000), // 10 minutes ago
lastAccess: new Date(now - 2 * 60 * 1000), // 2 minutes ago (not expired with 5 min timeout)
createdAt: new Date(now - 20 * 60 * 1000)
},
'expired-session': {
lastAccess: new Date(now - 40 * 60 * 1000), // 40 minutes ago (expired)
lastAccess: new Date(now - 10 * 60 * 1000), // 10 minutes ago (expired with 5 min timeout)
createdAt: new Date(now - 60 * 60 * 1000)
}
};
@@ -532,7 +534,7 @@ describe('HTTP Server Session Management', () => {
};
const metrics = (server as any).getSessionMetrics();
expect(metrics.totalSessions).toBe(2);
expect(metrics.activeSessions).toBe(2);
expect(metrics.expiredSessions).toBe(1);

View File

@@ -58,9 +58,9 @@ describe('TypeStructureService', () => {
});
describe('getAllStructures', () => {
it('should return all 22 type structures', () => {
it('should return all 23 type structures', () => {
const structures = TypeStructureService.getAllStructures();
expect(Object.keys(structures)).toHaveLength(22);
expect(Object.keys(structures)).toHaveLength(23);
});
it('should return a copy not a reference', () => {