fix: Docker container cleanup on session end (Issue #66)

- 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>

Dockerfile

@@ -69,6 +69,9 @@ ENV IS_DOCKER=true
 # Expose HTTP port
 EXPOSE 3000
 
+# Set stop signal to SIGTERM (default, but explicit is better)
+STOPSIGNAL SIGTERM
+
 # Health check
 HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
   CMD curl -f http://127.0.0.1:3000/health || exit 1

README.md

@@ -2,7 +2,7 @@
 
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![GitHub stars](https://img.shields.io/github/stars/czlonkowski/n8n-mcp?style=social)](https://github.com/czlonkowski/n8n-mcp)
-[![Version](https://img.shields.io/badge/version-2.7.19-blue.svg)](https://github.com/czlonkowski/n8n-mcp)
+[![Version](https://img.shields.io/badge/version-2.7.20-blue.svg)](https://github.com/czlonkowski/n8n-mcp)
 [![npm version](https://img.shields.io/npm/v/n8n-mcp.svg)](https://www.npmjs.com/package/n8n-mcp)
 [![n8n version](https://img.shields.io/badge/n8n-v1.102.4-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)
@@ -161,6 +161,7 @@ Add to Claude Desktop config:
         "run",
         "-i",
         "--rm",
+        "--init",
         "-e", "MCP_MODE=stdio",
         "-e", "LOG_LEVEL=error",
         "-e", "DISABLE_CONSOLE_OUTPUT=true",
@@ -181,6 +182,7 @@ Add to Claude Desktop config:
         "run",
         "-i",
         "--rm",
+        "--init",
         "-e", "MCP_MODE=stdio",
         "-e", "LOG_LEVEL=error",
         "-e", "DISABLE_CONSOLE_OUTPUT=true",
@@ -667,23 +669,20 @@ See [CHANGELOG.md](./docs/CHANGELOG.md) for full version history and recent chan
 
 ## ⚠️ Known Issues
 
-### Claude Desktop Container Duplication
+### Claude Desktop Container Management
-When using n8n-MCP with Claude Desktop in Docker mode, Claude Desktop may start the container twice during initialization. This is a known Claude Desktop bug ([modelcontextprotocol/servers#812](https://github.com/modelcontextprotocol/servers/issues/812)).
 
-**Symptoms:**
+#### Container Accumulation (Fixed in v2.7.20+)
-- Two identical containers running for the same MCP server
+Previous versions had an issue where containers would not properly clean up when Claude Desktop sessions ended. This has been fixed in v2.7.20+ with proper signal handling.
-- Container name conflicts if using `--name` parameter
-- Doubled resource usage
 
-**Workarounds:**
+**For best container lifecycle management:**
-1. **Avoid using --name parameter** - Let Docker assign random names:
+1. **Use the --init flag** (recommended) - Docker's init system ensures proper signal handling:
 ```json
 {
   "mcpServers": {
     "n8n-mcp": {
       "command": "docker",
       "args": [
-        "run", "-i", "--rm",
+        "run", "-i", "--rm", "--init",
         "ghcr.io/czlonkowski/n8n-mcp:latest"
       ]
     }
@@ -691,15 +690,11 @@ When using n8n-MCP with Claude Desktop in Docker mode, Claude Desktop may start
 }
 ```
 
-2. **Use HTTP mode instead** - Deploy n8n-mcp as a standalone HTTP server:
+2. **Ensure you're using v2.7.20 or later** - Check your version:
 ```bash
-docker compose up -d  # Start HTTP server
+docker run --rm ghcr.io/czlonkowski/n8n-mcp:latest --version
 ```
-Then connect via mcp-remote (see [HTTP Deployment Guide](./docs/HTTP_DEPLOYMENT.md))
 
-3. **Use Docker MCP Toolkit** - Better container management through Docker Desktop
-
-This issue does not affect the functionality of n8n-MCP itself, only the container management in Claude Desktop.
 
 ## 📦 License
 

docker/docker-entrypoint.sh

@@ -70,22 +70,12 @@ if [ "$(id -u)" = "0" ]; then
     if [ -d "/app/data" ]; then
         chown -R nodejs:nodejs /app/data
     fi
-    # Switch to nodejs user (using Alpine's native su)
+    # Switch to nodejs user with proper exec chain for signal propagation
-    exec su nodejs -c "$*"
+    exec su -s /bin/sh nodejs -c "exec $*"
 fi
 
-# Trap signals for graceful shutdown
+# Execute the main command directly with exec
-# In stdio mode, don't output anything to stdout as it breaks JSON-RPC
+# This ensures our Node.js process becomes PID 1 and receives signals directly
-if [ "$MCP_MODE" = "stdio" ]; then
-    # Silent trap - no output at all
-    trap 'kill -TERM $PID 2>/dev/null || true' TERM INT EXIT
-else
-    # In HTTP mode, output to stderr
-    trap 'echo "Shutting down..." >&2; kill -TERM $PID 2>/dev/null' TERM INT EXIT
-fi
-
-# Execute the main command in background
-# In stdio mode, use the wrapper for clean output
 if [ "$MCP_MODE" = "stdio" ]; then
     # Debug: Log to stderr to check if wrapper exists
     if [ "$DEBUG_DOCKER" = "true" ]; then
@@ -95,6 +85,7 @@ if [ "$MCP_MODE" = "stdio" ]; then
     
     if [ -f "/app/dist/mcp/stdio-wrapper.js" ]; then
         # Use the stdio wrapper for clean JSON-RPC output
+        # exec replaces the shell with node process as PID 1
         exec node /app/dist/mcp/stdio-wrapper.js
     else
         # Fallback: run with explicit environment

docs/CHANGELOG.md

@@ -5,6 +5,24 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.7.20] - 2025-07-18
+
+### Fixed
+- **Docker container cleanup on session end** (Issue #66)
+  - Fixed containers not responding to termination signals when Claude Desktop sessions end
+  - 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
+  - Recommended using `--init` flag in Docker run command for best signal handling
+
+### Documentation
+- Updated README with container lifecycle management best practices
+- Added `--init` flag to all Docker configuration examples
+- Added troubleshooting section for container accumulation issues
+
 ## [2.7.19] - 2025-07-18
 
 ### Fixed

package.json

@@ -1,6 +1,6 @@
 {
   "name": "n8n-mcp",
-  "version": "2.7.19",
+  "version": "2.7.20",
   "description": "Integration between n8n workflow automation and Model Context Protocol (MCP)",
   "main": "dist/index.js",
   "bin": {

src/mcp/server.ts

@@ -2031,4 +2031,18 @@ Full documentation is being prepared. For now, use get_node_essentials for confi
     // Keep the process alive and listening
     process.stdin.resume();
   }
+  
+  async shutdown(): Promise<void> {
+    logger.info('Shutting down MCP server...');
+    
+    // Close database connection if it exists
+    if (this.db) {
+      try {
+        await this.db.close();
+        logger.info('Database connection closed');
+      } catch (error) {
+        logger.error('Error closing database:', error);
+      }
+    }
+  }
 }

src/mcp/stdio-wrapper.ts

@@ -42,9 +42,11 @@ console.countReset = () => {};
 // Import and run the server AFTER suppressing output
 import { N8NDocumentationMCPServer } from './server';
 
+let server: N8NDocumentationMCPServer | null = null;
+
 async function main() {
   try {
-    const server = new N8NDocumentationMCPServer();
+    server = new N8NDocumentationMCPServer();
     await server.run();
   } catch (error) {
     // In case of fatal error, output to stderr only
@@ -64,4 +66,47 @@ process.on('unhandledRejection', (reason) => {
   process.exit(1);
 });
 
+// Handle termination signals for proper cleanup
+let isShuttingDown = false;
+
+async function shutdown(signal: string) {
+  if (isShuttingDown) return;
+  isShuttingDown = true;
+  
+  // Log to stderr only (not stdout which would corrupt JSON-RPC)
+  originalConsoleError(`Received ${signal}, shutting down gracefully...`);
+  
+  try {
+    // Shutdown the server if it exists
+    if (server) {
+      await server.shutdown();
+    }
+  } catch (error) {
+    originalConsoleError('Error during shutdown:', error);
+  }
+  
+  // Close stdin to signal we're done reading
+  process.stdin.pause();
+  process.stdin.destroy();
+  
+  // Exit with timeout to ensure we don't hang
+  setTimeout(() => {
+    process.exit(0);
+  }, 500).unref(); // unref() allows process to exit if this is the only thing keeping it alive
+  
+  // But also exit immediately if nothing else is pending
+  process.exit(0);
+}
+
+// Register signal handlers
+process.on('SIGTERM', () => void shutdown('SIGTERM'));
+process.on('SIGINT', () => void shutdown('SIGINT'));
+process.on('SIGHUP', () => void shutdown('SIGHUP'));
+
+// Also handle stdin close (when Claude Desktop closes the pipe)
+process.stdin.on('end', () => {
+  originalConsoleError('stdin closed, shutting down...');
+  void shutdown('STDIN_CLOSE');
+});
+
 main();