n8n workflows in manufacturing, you understand the stress of listening to {that a} course of failed and needing to dig via logs to search out the foundation trigger.
Person: Samir, your automation doesn’t work anymore, I didn’t obtain my notification!
Step one is to open your n8n interface and overview the final executions to establish the problems.
After a couple of minutes, you end up leaping between executions, evaluating timestamps and studying JSON errors to know the place issues broke.
What if an agent might let you know why your workflow failed at 3 AM with out you having to dig via the logs?
It’s attainable!
As an experiment, I related the n8n API, which gives entry to execution logs of my occasion, to an MCP server powered by Claude.
The result’s an AI assistant that may monitor workflows, analyse failures, and clarify what went improper in pure language.
On this article, I’ll stroll you thru the step-by-step technique of constructing this method.
The primary part will present an actual instance from my very own n8n occasion, the place a number of workflows failed through the night time.
We’ll use this case to see how the agent identifies points and explains their root causes.
Then, I’ll element how I related my n8n occasion’s API to the MCP server utilizing a webhook to allow Claude Desktop to fetch execution information for natural-language debugging.
The webhook consists of three features:
- Get Lively Workflows: which gives the listing of all lively workflows
- Get Final Executions: consists of details about the final n executions
- Get Executions Particulars (Standing = Error): particulars of failed executions formatted to assist root trigger analyses
You could find the entire tutorial, together with the n8n workflow template and the MCP server supply code, linked on this article.
Demonstration: Utilizing AI to Analyse Failed n8n Executions
Allow us to look collectively at one in all my n8n situations, which runs a number of workflows that fetch occasion info from totally different cities world wide.
These workflows assist enterprise and networking communities uncover attention-grabbing occasions to attend and be taught from.
To check the answer, I’ll begin by asking the agent to listing the lively workflows.
Step 1: What number of workflows are lively?
Based mostly on the query alone, Claude understood that it wanted to work together with the n8n-monitor device, which was constructed utilizing an MCP server.
From there, it mechanically chosen the corresponding operate, Get Lively Workflows, to retrieve the listing of lively automations from my n8n occasion.
That is the place you begin to sense the facility of the mannequin.
It mechanically categorised the workflows primarily based on their names
- 8 workflows to connect with fetch occasions from APIs and course of them
- 3 different workflows which might be work-in-progress, together with the one used to fetch the logs
This marks the start of the evaluation; all these insights might be utilised within the root trigger evaluation.
Step 2: Analyse the final n executions
At this stage, we will start asking Claude to retrieve the most recent executions for evaluation.
Because of the context offered within the doc-strings, which I’ll clarify within the subsequent part, Claude understood that it wanted to name the get workflow executions.
It would obtain a abstract of the executions, with the share of failures and the variety of workflows impacted by these failures.
{
"abstract": {
"totalExecutions": 25,
"successfulExecutions": 22,
"failedExecutions": 3,
"failureRate": "12.00%",
"successRate": "88.00%",
"totalWorkflowsExecuted": 7,
"workflowsWithFailures": 1
},
"executionModes": {
"webhook": 7,
"set off": 18
},
"timing": {
"averageExecutionTime": "15.75 seconds",
"maxExecutionTime": "107.18 seconds",
"minExecutionTime": "0.08 seconds",
"timeRange": {
"from": "2025-10-24T06:14:23.127Z",
"to": "2025-10-24T11:11:49.890Z"
}
},
[...]That is the very first thing it is going to share with you; it gives a transparent overview of the scenario.
Within the second a part of the outputs, you could find an in depth breakdown of the failures for every workflow impacted.
"failureAnalysis": {
"workflowsImpactedByFailures": [
"7uvA2XQPMB5l4kI5"
],
"failedExecutionsByWorkflow": {
"7uvA2XQPMB5l4kI5": {
"workflowId": "7uvA2XQPMB5l4kI5",
"failures": [
{
"id": "13691",
"startedAt": "2025-10-24T11:00:15.072Z",
"stoppedAt": "2025-10-24T11:00:15.508Z",
"mode": "trigger"
},
{
"id": "13683",
"startedAt": "2025-10-24T09:00:57.274Z",
"stoppedAt": "2025-10-24T09:00:57.979Z",
"mode": "trigger"
},
{
"id": "13677",
"startedAt": "2025-10-24T07:00:57.167Z",
"stoppedAt": "2025-10-24T07:00:57.685Z",
"mode": "trigger"
}
],
"failureCount": 3
}
},
"recentFailures": [
{
"id": "13691",
"workflowId": "7uvA2XQPMB5l4kI5",
"startedAt": "2025-10-24T11:00:15.072Z",
"mode": "trigger"
},
{
"id": "13683",
"workflowId": "7uvA2XQPMB5l4kI5",
"startedAt": "2025-10-24T09:00:57.274Z",
"mode": "trigger"
},
{
"id": "13677",
"workflowId": "7uvA2XQPMB5l4kI5",
"startedAt": "2025-10-24T07:00:57.167Z",
"mode": "trigger"
}
]
},As a consumer, you now have visibility into the impacted workflows, together with particulars of the failure occurrences.
For this particular case, the workflow “Bangkok Meetup” is triggered each hour.
What we might see is that we had points thrice (out of 5) over the last 5 hours.
Notice: We will ignore the final sentence because the agent doesn’t but have entry to the execution particulars.
The final part of the outputs consists of an evaluation of the general efficiency of the workflows.
"workflowPerformance": {
"allWorkflowMetrics": {
"CGvCrnUyGHgB7fi8": {
"workflowId": "CGvCrnUyGHgB7fi8",
"totalExecutions": 7,
"successfulExecutions": 7,
"failedExecutions": 0,
"successRate": "100.00%",
"failureRate": "0.00%",
"lastExecution": "2025-10-24T11:11:49.890Z",
"executionModes": {
"webhook": 7
}
},
[... other workflows ...]
,
"topProblematicWorkflows": [
{
"workflowId": "7uvA2XQPMB5l4kI5",
"totalExecutions": 5,
"successfulExecutions": 2,
"failedExecutions": 3,
"successRate": "40.00%",
"failureRate": "60.00%",
"lastExecution": "2025-10-24T11:00:15.072Z",
"executionModes": {
"trigger": 5
}
},
{
"workflowId": "CGvCrnUyGHgB7fi8",
"totalExecutions": 7,
"successfulExecutions": 7,
"failedExecutions": 0,
"successRate": "100.00%",
"failureRate": "0.00%",
"lastExecution": "2025-10-24T11:11:49.890Z",
"executionModes": {
"webhook": 7
}
},
[... other workflows ...]
}
]
}This detailed breakdown can assist you prioritise the upkeep in case you will have a number of workflows failing.
On this particular instance, I’ve solely a single failing workflow, which is the Ⓜ️ Bangkok Meetup.
What if I wish to know when points began?
Don’t fear, I’ve added a piece with the main points of the execution hour by hour.
"timeSeriesData": {
"2025-10-24T11:00": {
"whole": 5,
"success": 4,
"error": 1
},
"2025-10-24T10:00": {
"whole": 6,
"success": 6,
"error": 0
},
"2025-10-24T09:00": {
"whole": 3,
"success": 2,
"error": 1
},
"2025-10-24T08:00": {
"whole": 3,
"success": 3,
"error": 0
},
"2025-10-24T07:00": {
"whole": 3,
"success": 2,
"error": 1
},
"2025-10-24T06:00": {
"whole": 5,
"success": 5,
"error": 0
}
}You simply should let Claude create a pleasant visible just like the one you will have under.
Let me remind you right here that I didn’t present any suggestion of outcomes presentation to Claude; that is all from its personal initiative!
Spectacular, no?
Step 3: Root Trigger Evaluation
Now that we all know which workflows have points, we should always seek for the foundation trigger(s).
Claude ought to usually name the Get Error Executions operate to retrieve particulars of executions with failures.
On your info, the failure of this workflow is because of an error within the node JSON Tech that processes the output of the API name.
- Meetup Tech is sending an HTTP question to the Meetup API
- Processed by Consequence Tech Node
- JSON Tech is meant to remodel this output right into a remodeled JSON
Here’s what occurs when all the things goes properly.
Nonetheless, it could possibly occur that the API name generally fails and the JavaScript node receives an error, because the enter isn’t within the anticipated format.
Notice: This situation has been corrected in manufacturing since then (the code node is now extra sturdy), however I saved it right here for the demo.
Allow us to see if Claude can find the foundation trigger.
Right here is the output of the Get Error Executions operate.
{
"workflow_id": "7uvA2XQPMB5l4kI5",
"workflow_name": "Ⓜ️ Bangkok Meetup",
"error_count": 5,
"errors": [
{
"id": "13691",
"workflow_name": "Ⓜ️ Bangkok Meetup",
"status": "error",
"mode": "trigger",
"started_at": "2025-10-24T11:00:15.072Z",
"stopped_at": "2025-10-24T11:00:15.508Z",
"duration_seconds": 0.436,
"finished": false,
"retry_of": null,
"retry_success_id": null,
"error": {
"message": "A 'json' property isn't an object [item 0]",
"description": "Within the returned information, each key named 'json' should level to an object.",
"http_code": null,
"degree": "error",
"timestamp": null
},
"failed_node": {
"title": "JSON Tech",
"sort": "n8n-nodes-base.code",
"id": "dc46a767-55c8-48a1-a078-3d401ea6f43e",
"place": [
-768,
-1232
]
},
"set off": {}
},
[... 4 other errors ...]
],
"abstract": {
"total_errors": 5,
"error_patterns": {
"A 'json' property is not an object [item 0]": {
"rely": 5,
"executions": [
"13691",
"13683",
"13677",
"13660",
"13654"
]
}
},
"failed_nodes": {
"JSON Tech": 5
},
"time_range": {
"oldest": "2025-10-24T05:00:57.105Z",
"latest": "2025-10-24T11:00:15.072Z"
}
}
}Claude now has entry to the main points of the executions with the error message and the impacted nodes.
Within the response above, you possibly can see that Claude summarised the outputs of a number of executions in a single evaluation.
We all know now that:
- Errors occurred each hour besides at 08:00 am
- Every time, the identical node, referred to as “JSON Tech”, is impacted
- The error happens rapidly after the workflow is triggered
This descriptive evaluation is accomplished by the start of a diagnostic.
This assertion isn’t incorrect, as evidenced by the error message on the n8n UI.
Nonetheless, as a result of restricted context, Claude begins to supply suggestions to repair the workflow that aren’t right.
Along with the code correction, it gives an motion plan.
As I do know that the difficulty isn’t (solely) on the code node, I wished to information Claude within the root trigger evaluation.
It lastly challenged the preliminary proposal of the decision and commenced to share assumptions concerning the root trigger(s).
This begins to get nearer to the precise root trigger, offering sufficient insights for us to start out exploring the workflow.
The revised repair is now higher because it considers the chance that the difficulty comes from the node enter information.
For me, that is the very best I might anticipate from Claude, contemplating the restricted info that he has available.
Conclusion: Worth Proposition of This Instrument
This easy experiment demonstrates how an AI agent powered by Claude can prolong past primary monitoring to ship real operational worth.
Earlier than manually checking executions and logs, you possibly can first converse along with your automation system to ask what failed, why it failed, and obtain context-aware explanations inside seconds.
This won’t substitute you totally, however it could possibly speed up the foundation trigger evaluation course of.
Within the subsequent part, I’ll briefly introduce how I arrange the MCP Server to attach Claude Desktop to my occasion.
Constructing a neighborhood MCP Server to attach Claude Desktop to a FastAPI Microservice
To equip Claude with the three features out there within the webhook (Get Lively Workflows, Get Workflow Executions and Get Error Executions), I’ve applied an MCP Server.
On this part, I’ll briefly introduce the implementation, focusing solely on Get Lively Workflows and Get Workflows Executions, to display how I clarify the utilization of those instruments to Claude.
For a complete and detailed introduction to the answer, together with directions on deploy it on your machine, I invite you to observe this tutorial on my YouTube Channel.
Additionally, you will discover the MCP Server supply code and the n8n workflow of the webhook.
Create a Class to Question the Workflow
Earlier than inspecting arrange the three totally different instruments, let me introduce the utility class, which is outlined with all of the features wanted to work together with the webhook.
You could find it within the Python file: ./utils/n8n_monitory_sync.py
import logging
import os
from datetime import datetime, timedelta
from typing import Any, Dict, Optionally available
import requests
import traceback
logger = logging.getLogger(__name__)
class N8nMonitor:
"""Handler for n8n monitoring operations - synchronous model"""
def __init__(self):
self.webhook_url = os.getenv("N8N_WEBHOOK_URL", "")
self.timeout = 30Primarily, we retrieve the webhook URL from an surroundings variable and set a question timeout of 30 seconds.
The primary operate get_active_workflows is querying the webhook passing as a parameter: "motion": get_active_workflows".
def get_active_workflows(self) -> Dict[str, Any]:
"""Fetch all lively workflows from n8n"""
if not self.webhook_url:
logger.error("Atmosphere variable N8N_WEBHOOK_URL not configured")
return {"error": "N8N_WEBHOOK_URL surroundings variable not set"}
attempt:
logger.information("Fetching lively workflows from n8n")
response = requests.publish(
self.webhook_url,
json={"motion": "get_active_workflows"},
timeout=self.timeout
)
response.raise_for_status()
information = response.json()
logger.debug(f"Response sort: {sort(information)}")
# Record of all workflows
workflows = []
if isinstance(information, listing):
workflows = [item for item in data if isinstance(item, dict)]
if not workflows and information:
logger.error(f"Anticipated listing of dictionaries, obtained listing of {sort(information[0]).__name__}")
return {"error": "Webhook returned invalid information format"}
elif isinstance(information, dict):
if "information" in information:
workflows = information["data"]
else:
logger.error(f"Surprising dict response with keys: {listing(information.keys())} n {traceback.format_exc()}")
return {"error": "Surprising response format"}
else:
logger.error(f"Surprising response sort: {sort(information)} n {traceback.format_exc()}")
return {"error": f"Surprising response sort: {sort(information).__name__}"}
logger.information(f"Efficiently fetched {len(workflows)} lively workflows")
return {
"total_active": len(workflows),
"workflows": [
{
"id": wf.get("id", "unknown"),
"name": wf.get("name", "Unnamed"),
"created": wf.get("createdAt", ""),
"updated": wf.get("updatedAt", ""),
"archived": wf.get("isArchived", "false") == "true"
}
for wf in workflows
],
"abstract": {
"whole": len(workflows),
"names": [wf.get("name", "Unnamed") for wf in workflows]
}
}
besides requests.exceptions.RequestException as e:
logger.error(f"Error fetching workflows: {e} n {traceback.format_exc()}")
return {"error": f"Didn't fetch workflows: {str(e)} n {traceback.format_exc()}"}
besides Exception as e:
logger.error(f"Surprising error fetching workflows: {e} n {traceback.format_exc()}")
return {"error": f"Surprising error: {str(e)} n {traceback.format_exc()}"}I’ve added many checks, because the API generally fails to return the anticipated information format.
This resolution is extra sturdy, offering Claude with all the data to know why a question failed.
Now that the primary operate is roofed, we will concentrate on getting all of the final n executions with get_workflow_executions.
def get_workflow_executions(
self,
restrict: int = 50,
includes_kpis: bool = False,
) -> Dict[str, Any]:
"""Fetch workflow executions of the final 'restrict' executions with or with out KPIs """
if not self.webhook_url:
logger.error("Atmosphere variable N8N_WEBHOOK_URL not set")
return {"error": "N8N_WEBHOOK_URL surroundings variable not set"}
attempt:
logger.information(f"Fetching the final {restrict} executions")
payload = {
"motion": "get_workflow_executions",
"restrict": restrict
}
response = requests.publish(
self.webhook_url,
json=payload,
timeout=self.timeout
)
response.raise_for_status()
information = response.json()
if isinstance(information, listing) and len(information) > 0:
information = information[0]
logger.information("Efficiently fetched execution information")
if includes_kpis and isinstance(information, dict):
logger.information("Together with KPIs within the execution information")
if "abstract" in information:
abstract = information["summary"]
failure_rate = float(abstract.get("failureRate", "0").rstrip("%"))
information["insights"] = {
"health_status": "🟢 Wholesome" if failure_rate < 10 else
"🟡 Warning" if failure_rate < 25 else
"🔴 Important",
"message": f"{abstract.get('totalExecutions', 0)} executions with {abstract.get('failureRate', '0%')} failure fee"
}
return information
besides requests.exceptions.RequestException as e:
logger.error(f"HTTP error fetching executions: {e} n {traceback.format_exc()}")
return {"error": f"Didn't fetch executions: {str(e)}"}
besides Exception as e:
logger.error(f"Surprising error fetching executions: {e} n {traceback.format_exc()}")
return {"error": f"Surprising error: {str(e)}"}The one parameter right here is the quantity n of executions you wish to retrieve: "restrict": n.
The outputs embrace a abstract with a well being standing that’s generated by the code node Processing Audit. (more details in the tutorial)
The operate get_workflow_executions solely retrieves the outputs for formatting earlier than sending them to the agent.
Now that we have now outlined our core features, we will create the instruments to equip Claude through the MCP server.
Arrange an MCP Server with Instruments
Now it’s the time to create our MCP server with instruments and assets to equip (and train) Claude.
from mcp.server.fastmcp import FastMCP
import logging
from typing import Optionally available, Dict, Any
from utils.n8n_monitor_sync import N8nMonitor
logging.basicConfig(
degree=logging.INFO,
format='%(asctime)s - %(levelname)s - %(message)s',
handlers=[
logging.FileHandler("n8n_monitor.log"),
logging.StreamHandler()
]
)
logger = logging.getLogger(__name__)
mcp = FastMCP("n8n-monitor")
monitor = N8nMonitor()It’s a primary implementation utilizing FastMCP and importing n8n_monitor_sync.py with the features outlined within the earlier part.
# Useful resource for the agent (Samir: replace it every time you add a device)
@mcp.useful resource("n8n://assist")
def get_help() -> str:
"""Get assist documentation for the n8n monitoring instruments"""
return """
📊 N8N MONITORING TOOLS
=======================
WORKFLOW MONITORING:
• get_active_workflows()
Record all lively workflows with names and IDs
EXECUTION TRACKING:
• get_workflow_executions(restrict=50, include_kpis=True)
Get execution logs with detailed KPIs
- restrict: Variety of latest executions to retrieve (1-100)
- include_kpis: Calculate efficiency metrics
ERROR DEBUGGING:
• get_error_executions(workflow_id)
Retrieve detailed error info for a selected workflow
- Returns final 5 errors with complete debugging information
- Reveals error messages, failed nodes, set off information
- Identifies error patterns and problematic nodes
- Contains HTTP codes, error ranges, and timing information
HEALTH REPORTING:
• get_workflow_health_report(restrict=50)
Generate complete well being evaluation primarily based on latest executions
- Identifies problematic workflows
- Reveals success/failure charges
- Supplies execution timing metrics
KEY METRICS PROVIDED:
• Whole executions
• Success/failure charges
• Execution occasions (avg, min, max)
• Workflows with failures
• Execution modes (guide, set off, built-in)
• Error patterns and frequencies
• Failed node identification
HEALTH STATUS INDICATORS:
• 🟢 Wholesome: <10% failure fee
• 🟡 Warning: 10-25% failure fee
• 🔴 Important: >25% failure fee
USAGE EXAMPLES:
- "Present me all lively workflows"
- "What workflows have been failing?"
- "Generate a well being report for my n8n occasion"
- "Present execution metrics for the final 48 hours"
- "Debug errors in workflow CGvCrnUyGHgB7fi8"
- "What's inflicting failures in my information processing workflow?"
DEBUGGING WORKFLOW:
1. Use get_workflow_executions() to establish problematic workflows
2. Use get_error_executions() for detailed error evaluation
3. Examine error patterns to establish recurring points
4. Evaluate failed node particulars and set off information
5. Use workflow_id and execution_id for focused fixes
"""Because the device is complicated to apprehend, we embrace a immediate, within the type of an MCP useful resource, to summarise the target and options of the n8n workflow related through webhook.
Now we will outline the primary device to get all of the lively workflows.
@mcp.device()
def get_active_workflows() -> Dict[str, Any]:
"""
Get all lively workflows within the n8n occasion.
Returns:
Dictionary with listing of lively workflows and their particulars
"""
attempt:
logger.information("Fetching lively workflows")
outcome = monitor.get_active_workflows()
if "error" in outcome:
logger.error(f"Didn't get workflows: {outcome['error']}")
else:
logger.information(f"Discovered {outcome.get('total_active', 0)} lively workflows")
return outcome
besides Exception as e:
logger.error(f"Surprising error: {str(e)}")
return {"error": str(e)}The docstring, used to clarify to the MCP server use the device, is comparatively transient, as there are not any enter parameters for get_active_workflows().
Allow us to do the identical for the second device to retrieve the final n executions.
@mcp.device()
def get_workflow_executions(
restrict: int = 50,
include_kpis: bool = True
) -> Dict[str, Any]:
"""
Get workflow execution logs and KPIs for the final N executions.
Args:
restrict: Variety of executions to retrieve (default: 50)
include_kpis: Embrace calculated KPIs (default: true)
Returns:
Dictionary with execution information and KPIs
"""
attempt:
logger.information(f"Fetching the final {restrict} executions")
outcome = monitor.get_workflow_executions(
restrict=restrict,
includes_kpis=include_kpis
)
if "error" in outcome:
logger.error(f"Didn't get executions: {outcome['error']}")
else:
if "abstract" in outcome:
abstract = outcome["summary"]
logger.information(f"Executions: {abstract.get('totalExecutions', 0)}, "
f"Failure fee: {abstract.get('failureRate', 'N/A')}")
return outcome
besides Exception as e:
logger.error(f"Surprising error: {str(e)}")
return {"error": str(e)}In contrast to the earlier device, we have to specify the enter information with the default worth.
We now have now outfitted Claude with these two instruments that can be utilized as within the instance offered within the earlier part.
What’s subsequent? Deploy it in your machine!
As I wished to maintain this text brief, I’ll solely introduce these two instruments.
For the remainder of the functionalities, I invite you to observe this complete tutorial on my YouTube channel.
I embrace step-by-step explanations on deploy this in your machine with an in depth overview of the supply code shared on my GitHub (MCP Server) and n8n profile (workflow).
Conclusion
That is only the start!
We will contemplate this as model 1.0 of what can develop into an excellent agent to handle your n8n workflows.
What do I imply by this?
There’s a large potential for bettering this resolution, particularly for the foundation trigger evaluation by:
- Offering extra context to the agent utilizing the sticky notes contained in the workflows
- Displaying how good inputs and outputs look with analysis nodes to assist Claude carry out hole analyses
- Exploiting the opposite endpoints of the n8n API for extra correct analyses
Nonetheless, I don’t suppose I can, as a full-time startup founder and CEO, develop such a complete device by myself.
Due to this fact, I wished to share that with the In direction of Information Science and n8n group as an open-source resolution out there on my GitHub profile.
Want inspiration to start out automating with n8n?
On this weblog, I’ve printed a number of articles to share examples of workflow automations we have now applied for small, medium and enormous operations.
The main focus was primarily on logistics and provide chain operations with actual case research:
I even have a complete playlist on my YouTube Channel, Provide Science, with greater than 15 tutorials.
You’ll be able to comply with these tutorials to deploy the workflows I share on my n8n creator profile (linked within the descriptions) that cowl:
- Course of Automation for Logistics and Provide Chain
- AI-Powered Workflows for Content material Creation
- Productiveness and Language Studying
Be happy to share your questions within the remark sections of the movies.
Different examples of MCP Server Implementation
This isn’t my first implementation of MCP servers.
In one other experiment, I related Claude Desktop with a Provide-Chain Community Optimisation device.
On this instance, the n8n workflow is changed by a FastAPI microservice internet hosting a linear programming algorithm.
The target is to find out the optimum set of factories to supply and ship merchandise to market on the lowest price and with the smallest environmental footprint.
In the sort of train, Claude is doing an incredible job of synthesising and presenting outcomes.
For extra info, take a look at this In direction of Information Science Article.
About Me
Let’s join on Linkedin and Twitter. I’m a Provide Chain Engineer who makes use of information analytics to enhance logistics operations and scale back prices.
For consulting or recommendation on analytics and sustainable provide chain transformation, be at liberty to contact me through Logigreen Consulting.
If you’re enthusiastic about Information Analytics and Provide Chain, have a look at my web site.
