Mcp triger

Questions
1

Hi everyone,
I’m facing a blocking issue with the new MCP (Model Context Protocol) trigger in a self-hosted n8n environment. While the handshake works, calling a tool causes the Webhook pod to hang or crash with a specific error.
My Setup:
n8n Version: 1.123.9
Deployment: Kubernetes (K8s)
Mode: Queue Mode (Separate Main, Webhook, and Worker pods)
Database: Postgres
Redis: Valkey (Redis-compatible)
The Issue:
Handshake works: Running an initialize or tools/list request via CURL returns the correct capabilities and tool definitions.
Tool Call hangs: When calling tools/call (either via CURL or Cursor), the request hangs indefinitely.
Redis State: Checking Redis with llen bull:n8n-queue:wait or active returns 0. It seems the job is never pushed to the queue.
Webhook Logs: The Webhook pod throws the following error: Error in handling webhook request: Cannot read properties of undefined (reading ‘node’)
Troubleshooting already done:
Encryption Key: Verified that N8N_ENCRYPTION_KEY is identical across all pods (Main, Webhook, and Worker).
Execution Mode: Verified EXECUTIONS_MODE=queue is set for all components.
Workflow State: The workflow is Active. I’ve tried toggling it Off/On and re-saving.
Paths: Verified the URL path matches the MCP trigger configuration exactly (no trailing slashes).
Task Runners: Tried both N8N_RUNNERS_ENABLED=true and false on the Worker, but since the job doesn’t reach Redis, it seems the issue is at the Webhook/Main level.
It feels like the Webhook pod identifies the incoming request but fails to look up the workflow/node details from the database, leading to the undefined (reading ‘node’) error.
Has anyone encountered this specific error with MCP in a distributed/queue setup? Any help would be greatly appreciated!
If your question hasn’t been asked before, please follow the template below. Skip the questions that are not relevant to you. →

Describe the problem/error/question

What is the error message (if any)?

Please share your workflow

(Select the nodes on your canvas and use the keyboard shortcuts CMD+C/CTRL+C and CMD+V/CTRL+V to copy and paste the workflow.)

Share the output returned by the last node

Information on your n8n setup

  • n8n version:
  • Database (default: SQLite):
  • n8n EXECUTIONS_PROCESS setting (default: own, main):
  • Running n8n via (Docker, npm, n8n cloud, desktop app):
  • Operating system:
2

Hi, @Refael_Ben_Chemo !

This is not a configuration issue. The MCP Trigger currently has limitations in distributed / queue-mode setups.

What’s happening is that the handshake works, but on tools/call the Webhook pod fails to resolve the workflow/node from the database, leading to the Cannot read properties of undefined (reading ‘node’) error. Because of this, the execution is never created and never reaches Redis, which is why the queue stays empty.

At the moment, the MCP Trigger works reliably only in single-instance setups (no queue mode, main + webhook together). The recommended approach is to run MCP without queue mode, or use a standard Webhook + AI Agent flow until MCP support for distributed environments is improved.