DEV Community: 松宮亮介

5 n8n Automation Workflows That Save Me 10+ Hours Every Week

松宮亮介 — Thu, 02 Apr 2026 08:18:11 +0000

I replaced Zapier, three Python scripts, and a VA with five n8n workflows. Here's exactly how each one works, with importable JSON.

Why n8n Over Zapier or Make?

Before diving into the workflows, let me address the obvious question: why n8n?

I've used Zapier (3 years), Make/Integromat (1 year), and n8n (2 years). Here's my honest comparison:

Feature	Zapier	Make	n8n
Pricing	$29+/mo for 750 tasks	$9+/mo for 10K ops	Free (self-hosted)
Self-hosting	No	No	Yes
Code nodes	Limited	JavaScript only	JavaScript + Python
AI integrations	Basic	Basic	Native Claude, GPT, etc.
Complex logic	Painful	Better	Full programming power
Data privacy	Cloud only	Cloud only	Your server, your data

The dealbreaker for me was cost. Running 50,000+ operations/month on Zapier would cost $100+. On n8n self-hosted, it costs the $6/month VPS fee — regardless of volume.

Let's look at the five workflows that save me the most time.

Workflow 1: Automated Lead Management Pipeline

Time saved: ~3 hours/week

When a new lead fills out my contact form (Google Forms), this workflow automatically:

Adds the lead to a Google Sheet with a timestamp
Sends a personalized welcome email
Posts a notification to my team's Slack channel
Creates a follow-up reminder for 3 days later

How It Works

The workflow starts with a Google Forms Trigger that fires whenever someone submits the form. A Google Sheets node appends the lead data to a tracking spreadsheet. Then it branches into three parallel paths: email, Slack notification, and a delayed follow-up.

Key Nodes

{
  "name": "Lead Pipeline",
  "nodes": [
    {
      "parameters": {
        "formId": "YOUR_FORM_ID"
      },
      "name": "New Form Submission",
      "type": "n8n-nodes-base.googleFormsTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "operation": "append",
        "sheetId": "YOUR_SHEET_ID",
        "range": "Leads!A:F",
        "columns": "timestamp,name,email,company,source,status"
      },
      "name": "Add to Sheet",
      "type": "n8n-nodes-base.googleSheets",
      "typeVersion": 4,
      "position": [450, 300]
    },
    {
      "parameters": {
        "resource": "message",
        "operation": "send",
        "to": "={{ $json.email }}",
        "subject": "Thanks for reaching out, {{ $json.name }}!",
        "emailType": "html",
        "message": "<h2>Hi {{ $json.name }},</h2><p>Thanks for your interest. I'll review your request and get back to you within 24 hours.</p>"
      },
      "name": "Welcome Email",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 2,
      "position": [650, 200]
    },
    {
      "parameters": {
        "channel": "#leads",
        "text": "New lead from {{ $json.name }} ({{ $json.company }})"
      },
      "name": "Slack Alert",
      "type": "n8n-nodes-base.slack",
      "typeVersion": 2,
      "position": [650, 400]
    }
  ]
}

Pro Tip: Lead Scoring

Add a Code node after the Google Sheets node to score leads automatically:

const lead = .first().json;
let score = 0;

// Score based on company size indicators
if (lead.company && lead.company.length > 0) score += 20;
if (lead.source === 'referral') score += 30;
if (lead.source === 'google') score += 10;

// Add score to the data
return [{ json: { ...lead, leadScore: score, priority: score >= 30 ? 'high' : 'normal' } }];

Workflow 2: Social Media Auto-Scheduling

Time saved: ~2 hours/week

I write content in a Google Doc once a week. This workflow takes that content and schedules posts across multiple platforms.

How It Works

A Schedule Trigger runs every Monday at 9 AM. It reads a Google Sheet containing pre-written posts (with columns for platform, content, and scheduled time). For each row marked as "pending," it posts to the appropriate platform via API and updates the status to "posted."

Key Nodes

{
  "name": "Social Media Scheduler",
  "nodes": [
    {
      "parameters": {
        "rule": {
          "interval": [{ "triggerAtHour": 9, "triggerAtDay": 1 }]
        }
      },
      "name": "Monday 9AM",
      "type": "n8n-nodes-base.scheduleTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "operation": "read",
        "sheetId": "YOUR_SHEET_ID",
        "range": "Posts!A:E"
      },
      "name": "Get Pending Posts",
      "type": "n8n-nodes-base.googleSheets",
      "typeVersion": 4,
      "position": [450, 300]
    },
    {
      "parameters": {
        "conditions": {
          "string": [
            {
              "value1": "={{ $json.platform }}",
              "value2": "twitter"
            }
          ]
        }
      },
      "name": "Route by Platform",
      "type": "n8n-nodes-base.switch",
      "typeVersion": 3,
      "position": [650, 300]
    }
  ]
}

Platform-Specific Posting

For Twitter/X, use the HTTP Request node with the Twitter API v2:

// Code node to format for Twitter (280 char limit)
const post = .first().json;
let tweet = post.content;

if (tweet.length > 280) {
  tweet = tweet.substring(0, 277) + '...';
}

return [{ json: { text: tweet, originalId: post.id } }];

For LinkedIn, use the HTTP Request node with the LinkedIn Share API. The key is formatting the post body correctly:

{
  "author": "urn:li:person:YOUR_PERSON_ID",
  "lifecycleState": "PUBLISHED",
  "specificContent": {
    "com.linkedin.ugc.ShareContent": {
      "shareCommentary": { "text": "{{ $json.content }}" },
      "shareMediaCategory": "NONE"
    }
  },
  "visibility": {
    "com.linkedin.ugc.MemberNetworkVisibility": "PUBLIC"
  }
}

Workflow 3: AI-Powered Email Draft Generator

Time saved: ~2.5 hours/week

This is my favorite workflow. It monitors my Gmail for emails that need replies, uses Claude API to draft responses, and saves the drafts for my review.

How It Works

A Gmail Trigger fires on new emails matching a specific label (I use "needs-reply"). The email content is sent to Claude API with a system prompt that includes my writing style and common response patterns. Claude generates a draft reply, which is saved as a Gmail draft attached to the original thread.

Key Nodes

{
  "name": "AI Email Drafter",
  "nodes": [
    {
      "parameters": {
        "pollTimes": { "item": [{ "mode": "everyMinute", "minute": 5 }] },
        "filters": { "labelIds": ["Label_needs_reply"] }
      },
      "name": "New Email",
      "type": "n8n-nodes-base.gmailTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "method": "POST",
        "url": "https://api.anthropic.com/v1/messages",
        "sendHeaders": true,
        "headerParameters": {
          "parameters": [
            { "name": "x-api-key", "value": "={{ $env.ANTHROPIC_API_KEY }}" },
            { "name": "anthropic-version", "value": "2023-06-01" },
            { "name": "content-type", "value": "application/json" }
          ]
        },
        "sendBody": true,
        "specifyBody": "json"
      },
      "name": "Claude Draft",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4.2,
      "position": [500, 300]
    },
    {
      "parameters": {
        "resource": "draft",
        "operation": "create",
        "subject": "Re: {{ $('New Email').item.json.subject }}",
        "message": "={{ $json.content[0].text }}",
        "options": {
          "threadId": "={{ $('New Email').item.json.threadId }}"
        }
      },
      "name": "Save Draft",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 2,
      "position": [750, 300]
    }
  ]
}

Why Drafts, Not Auto-Send?

I intentionally save as drafts rather than auto-sending. AI-generated emails still need a human eye for:

Tone — Claude is good but not perfect at matching your voice
Accuracy — Factual claims should be verified
Relationships — Some emails need a more personal touch

The time savings come from having a 90%-ready draft instead of starting from a blank page. I typically spend 10-20 seconds reviewing each draft versus 3-5 minutes writing from scratch.

Workflow 4: Web Scraping + Data Processing + Report Generation

Time saved: ~2 hours/week

Every Friday, this workflow scrapes competitor pricing pages, processes the data, generates a comparison report, and emails it to me.

How It Works

A Schedule Trigger fires every Friday at 6 AM. For each competitor URL in a predefined list, an HTTP Request node fetches the page HTML. A Code node extracts pricing data using regex patterns. Finally, Claude API generates a natural-language summary report, which is sent via email.

Key Nodes

{
  "name": "Competitor Price Monitor",
  "nodes": [
    {
      "parameters": {
        "rule": {
          "interval": [{ "triggerAtHour": 6, "triggerAtDay": 5 }]
        }
      },
      "name": "Friday 6AM",
      "type": "n8n-nodes-base.scheduleTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "jsCode": "const urls = [
  { name: 'Competitor A', url: 'https://competitor-a.com/pricing' },
  { name: 'Competitor B', url: 'https://competitor-b.com/pricing' }
];
return urls.map(u => ({ json: u }));"
      },
      "name": "URL List",
      "type": "n8n-nodes-base.code",
      "typeVersion": 2,
      "position": [450, 300]
    },
    {
      "parameters": {
        "url": "={{ $json.url }}",
        "options": { "timeout": 10000 }
      },
      "name": "Fetch Page",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4.2,
      "position": [650, 300]
    }
  ]
}

Data Extraction Code Node

const html = .first().json.data;
const name = .first().json.name;

// Extract pricing data (customize regex for each site)
const priceRegex = /$(d+(?:.d{2})?)s*(?:/s*mo|pers*month)/gi;
const prices = [];
let match;

while ((match = priceRegex.exec(html)) !== null) {
  prices.push(parseFloat(match[1]));
}

return [{
  json: {
    competitor: name,
    prices: prices,
    scrapedAt: new Date().toISOString()
  }
}];

Report Generation with Claude

After collecting all competitor data, merge the results and send them to Claude for analysis:

const allData = .all().map(item => item.json);

const prompt = "Analyze this competitor pricing data and write a brief report 
highlighting key changes, trends, and actionable insights:

"
  + JSON.stringify(allData, null, 2);

return [{ json: { prompt } }];

Workflow 5: Telegram Bot for Task Management

Time saved: ~1.5 hours/week

This workflow turns Telegram into a lightweight task manager. I send tasks via Telegram, and they are automatically organized, prioritized, and synced with a Google Sheet.

Commands

Command	Action
`/add Buy groceries`	Adds a new task
`/list`	Shows all pending tasks
`/done 3`	Marks task #3 as complete
`/priority 2 high`	Sets task #2 to high priority

How It Works

A Telegram Trigger receives messages. A Switch node routes based on the command prefix. Each branch handles the specific operation — adding, listing, completing, or updating tasks in a Google Sheet.

Key Nodes

{
  "name": "Telegram Task Manager",
  "nodes": [
    {
      "parameters": {
        "updates": ["message"]
      },
      "name": "Telegram Trigger",
      "type": "n8n-nodes-base.telegramTrigger",
      "typeVersion": 1.1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "rules": {
          "rules": [
            { "value": "/add", "output": 0 },
            { "value": "/list", "output": 1 },
            { "value": "/done", "output": 2 },
            { "value": "/priority", "output": 3 }
          ]
        },
        "dataType": "string",
        "value1": "={{ $json.message.text.split(' ')[0] }}"
      },
      "name": "Command Router",
      "type": "n8n-nodes-base.switch",
      "typeVersion": 3,
      "position": [450, 300]
    }
  ]
}

Add Task Handler

const text = .message.text;
const taskText = text.replace(/^/adds+/, '');
const chatId = .message.chat.id;

if (!taskText) {
  return [{ json: { chatId, response: 'Usage: /add <task description>' } }];
}

const task = {
  id: Date.now(),
  text: taskText,
  status: 'pending',
  priority: 'normal',
  createdAt: new Date().toISOString(),
  chatId: chatId
};

return [{ json: { ...task, response: 'Task added: ' + taskText } }];

List Tasks Handler

const tasks = .all().map(item => item.json);
const pending = tasks.filter(t => t.status === 'pending');

if (pending.length === 0) {
  return [{ json: { response: 'No pending tasks. Use /add to create one.' } }];
}

const list = pending.map((t, i) => {
  const pri = t.priority === 'high' ? '[!]' : '   ';
  return (i + 1) + '. ' + pri + ' ' + t.text;
}).join('
');

return [{ json: { response: 'Pending tasks:

' + list } }];

Performance Tips for n8n Workflows

After running these workflows for over a year, here are my top optimization tips:

1. Use Static Data for Caching

n8n's () persists data between executions without needing a database:

const cache = ('global');
const cacheKey = 'lastResult_' + someId;
const cacheMaxAge = 3600000; // 1 hour

if (cache[cacheKey] && (Date.now() - cache[cacheKey].timestamp) < cacheMaxAge) {
  return [{ json: cache[cacheKey].data }];
}

// ... fetch fresh data ...
cache[cacheKey] = { data: freshData, timestamp: Date.now() };

2. Error Handling on Every API Call

Always add an error handler. API calls fail more often than you think — rate limits, network issues, expired tokens:

try {
  const result = .first().json;
  if (result.error) throw new Error(result.error.message);
  return [{ json: result }];
} catch (err) {
  return [{ json: { error: true, message: err.message, fallback: 'Default response' } }];
}

3. Batch Operations Where Possible

Instead of making 100 individual API calls, batch them:

const items = .all().map(item => item.json);
const batchSize = 10;
const batches = [];

for (let i = 0; i < items.length; i += batchSize) {
  batches.push(items.slice(i, i + batchSize));
}

return batches.map(batch => ({ json: { batch } }));

4. Monitor Execution Times

n8n shows execution time per node. If a workflow takes more than 30 seconds, look for:

Unnecessary API calls that could be cached
Sequential operations that could run in parallel (use the Split In Batches node)
Large payloads being passed between nodes (extract only what you need)

5. Use Environment Variables

Never hardcode API keys or sensitive data in workflows:

// In n8n, reference environment variables like this:
const apiKey = .ANTHROPIC_API_KEY;
const webhookUrl = .SLACK_WEBHOOK_URL;

Set them in your Docker Compose or .env file, not in the workflow JSON.

Wrapping Up

These five workflows have fundamentally changed how I work. The initial setup takes a few hours each, but the compounding time savings are enormous:

Workflow	Weekly Time Saved
Lead Management Pipeline	3 hours
Social Media Scheduling	2 hours
AI Email Draft Generator	2.5 hours
Web Scraping + Reports	2 hours
Telegram Task Manager	1.5 hours
Total	11 hours/week

That is over 500 hours per year redirected from repetitive tasks to actual creative and strategic work.

The best part? Every one of these workflows runs on a single $6/month VPS. No SaaS subscriptions, no per-execution fees, no vendor lock-in.

Start with one workflow — whichever solves your biggest pain point — and build from there. The JSON snippets above are ready to import into n8n and customize for your needs.

If you want to see how I combine all of these with Claude API for even more automation, check out my previous article: Building an AI-Powered Telegram Bot with Claude API and n8n.

Ryosuke Matsumiya — AI Automation Engineer. Building OpenClaw, an AI-powered Telegram agent. Passionate about n8n, Claude API, and workflow automation.

Building an AI-Powered Telegram Bot with Claude API and n8n: A Complete Guide

松宮亮介 — Thu, 02 Apr 2026 08:17:47 +0000

Build a production-ready AI chatbot on Telegram using Anthropic's Claude API and n8n in under 30 minutes, with zero traditional backend code.

Why Claude API + Telegram + n8n?

If you've ever wanted to deploy an AI assistant that real users can interact with instantly, you've probably faced a familiar dilemma: write a full backend from scratch, or cobble together fragile scripts that break at 3 AM.

There's a better way. By combining three tools — Claude API for intelligence, Telegram for the user interface, and n8n for orchestration — you get a production-grade chatbot with:

Claude 3.5 Sonnet / Claude 4 as the reasoning engine — far more nuanced than GPT for following complex instructions
Telegram as a zero-friction UI that 900M+ users already have installed
n8n as the glue layer — visual workflow automation that you self-host, so there are no per-execution fees like Zapier

I've been running this exact stack for OpenClaw, an AI-powered Telegram agent, and it handles thousands of messages daily without a single line of Express.js. Let me show you how to build it.

Prerequisites

Before we start, make sure you have:

Requirement	Details
Node.js	v18+ (required for n8n)
n8n	Self-hosted or n8n Cloud account
Anthropic API Key	Get one at console.anthropic.com
Telegram Account	To create a bot via BotFather

Estimated cost: ~$0.003 per message with Claude 3.5 Sonnet. For most personal-use bots, you'll spend less than $5/month.

Step 1: Create Your Telegram Bot with BotFather

Open Telegram and search for @botfather. This is Telegram's official bot for creating bots.

/start
/newbot

BotFather will ask two questions:

Bot name: The display name (e.g., "My Claude Assistant")
Bot username: Must end in bot (e.g., my_claude_assistant_bot)

You'll receive a Bot Token that looks like this:

7123456789:AAH_your-token-string-here

Save this token — you'll need it in Step 3.

Optional but Recommended: Configure Bot Settings

/setdescription - Add a description for your bot
/setabouttext   - Set the "About" section
/setuserpic     - Upload a profile picture
/setcommands    - Define command menu

Here's a useful command list to set via /setcommands:

start - Start the bot
help - Show available commands
clear - Reset conversation context

Step 2: Install and Set Up n8n

Option A: Self-Hosted (Recommended for Production)

# Install n8n globally
npm install -g n8n

# Or use Docker (preferred for production)
docker run -d \
  --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  -e WEBHOOK_URL=https://your-domain.com/ \
  n8nio/n8n

Important: For Telegram webhooks to work, n8n must be accessible via HTTPS on a public URL. If you're developing locally, use a tunnel service like ngrok or cloudflared.

# Quick tunnel for local development
npx ngrok http 5678

Option B: n8n Cloud

Sign up at app.n8n.cloud. The free tier gives you 2,500 executions/month — more than enough for testing. Webhooks work out of the box, no tunnel required.

Verify Installation

Open http://localhost:5678 (or your cloud URL). You should see the n8n canvas editor. Create a new workflow and name it "Claude Telegram Bot".

Step 3: Configure the Telegram Trigger Node

In your new workflow, click "Add first step" and search for Telegram.

Select "Telegram Trigger" — this node listens for incoming messages to your bot.

Configure Credentials

Click "Create New Credential"
Paste your Bot Token from Step 1
Click "Save"

Trigger Settings

Setting	Value
Updates	`message`
Additional Fields	Leave default

The node configuration JSON:

{
  "parameters": {
    "updates": ["message"]
  },
  "name": "Telegram Trigger",
  "type": "n8n-nodes-base.telegramTrigger",
  "typeVersion": 1.1,
  "position": [250, 300],
  "credentials": {
    "telegramApi": {
      "id": "YOUR_CREDENTIAL_ID",
      "name": "Telegram Bot"
    }
  }
}

Test the Trigger

Click "Listen for test event" in n8n
Send any message to your bot on Telegram
You should see the incoming message data appear in n8n

The payload includes message.text, message.chat.id, message.from.first_name, and more. We'll use these fields in subsequent steps.

Step 4: Connect the Claude API (Anthropic Node)

Add a new node after the Telegram Trigger. Search for "Anthropic" or "HTTP Request".

Option A: Using the Built-in Anthropic Node (n8n v1.30+)

n8n v1.30+ ships with a native Anthropic node. Configure it as follows:

Credential: Add your Anthropic API key
Model: claude-sonnet-4-20250514 (best balance of speed, quality, and cost)
Prompt: Use an expression to inject the user's Telegram message:

{{ $json.message.text }}

Option B: Using HTTP Request Node (Any n8n Version)

If your n8n version doesn't have the native Anthropic node, the HTTP Request node works perfectly:

{
  "parameters": {
    "method": "POST",
    "url": "https://api.anthropic.com/v1/messages",
    "authentication": "genericCredentialType",
    "genericAuthType": "httpHeaderAuth",
    "sendHeaders": true,
    "headerParameters": {
      "parameters": [
        {
          "name": "x-api-key",
          "value": "YOUR_ANTHROPIC_API_KEY"
        },
        {
          "name": "anthropic-version",
          "value": "2023-06-01"
        },
        {
          "name": "content-type",
          "value": "application/json"
        }
      ]
    },
    "sendBody": true,
    "specifyBody": "json",
    "jsonBody": "={{ JSON.stringify({ model: 'claude-sonnet-4-20250514', max_tokens: 1024, system: 'You are a helpful AI assistant on Telegram. Keep responses concise and formatted for mobile reading. Use short paragraphs and bullet points when appropriate.', messages: [{ role: 'user', content: $json.message.text }] }) }}"
  },
  "name": "Claude API",
  "type": "n8n-nodes-base.httpRequest",
  "typeVersion": 4.2,
  "position": [500, 300]
}

System Prompt Best Practices for Telegram Bots

The system prompt determines your bot's personality and behavior. Here's one that works well:

You are a helpful AI assistant on Telegram. Follow these rules:
1. Keep responses under 4096 characters (Telegram's message limit).
2. Use short paragraphs — mobile screens are small.
3. Use bullet points and numbered lists for structured information.
4. If a question is ambiguous, give the most likely answer first,
   then briefly mention alternatives.
5. Format code blocks with triple backticks and language identifiers.

Step 5: Send the Response Back to Telegram

Add a "Telegram" node (the action node, not the Trigger) after the Claude API node.

Configuration

Setting	Value
Resource	Send Message
Chat ID	`{{ $('Telegram Trigger').item.json.message.chat.id }}`
Text	Depends on Step 4 method (see below)
Parse Mode	Markdown

For the Text field:

If using the native Anthropic node:

{{ $json.message.content[0].text }}

If using the HTTP Request node:

{{ $json.content[0].text }}

Handle Long Responses

Telegram enforces a 4096-character limit per message. Add a Code node between Claude API and the Telegram send node to split long responses:

const text = $input.first().json.content[0].text;
const chatId = $('Telegram Trigger').first().json.message.chat.id;
const maxLength = 4000; // Leave buffer for Markdown formatting

if (text.length <= maxLength) {
  return [{ json: { chatId, text } }];
}

// Split into chunks at natural paragraph boundaries
const chunks = [];
let remaining = text;

while (remaining.length > 0) {
  if (remaining.length <= maxLength) {
    chunks.push(remaining);
    break;
  }

  let splitIndex = remaining.lastIndexOf('\n\n', maxLength);
  if (splitIndex === -1) splitIndex = remaining.lastIndexOf('\n', maxLength);
  if (splitIndex === -1) splitIndex = remaining.lastIndexOf(' ', maxLength);
  if (splitIndex === -1) splitIndex = maxLength;

  chunks.push(remaining.substring(0, splitIndex));
  remaining = remaining.substring(splitIndex).trimStart();
}

return chunks.map(chunk => ({ json: { chatId, text: chunk } }));

Then update the Telegram send node to use {{ $json.text }} for the text and {{ $json.chatId }} for the chat ID.

Step 6: Error Handling and Fallback

Production bots need graceful error handling. Here's how to make yours resilient.

Add an Error Trigger Workflow

Create a separate workflow called "Bot Error Handler":

Add an Error Trigger node
Connect it to a Telegram Send Message node
Configure the error message:

Sorry, I encountered an error processing your request.
Please try again in a moment.

If this persists, use /clear to reset the conversation.

Rate Limiting

Claude API has rate limits, and users sometimes spam messages. Add a Code node before the API call:

const chatId = $json.message.chat.id;
const now = Date.now();
const rateLimitMs = 2000; // Minimum 2 seconds between messages

// Use n8n's static data for cross-execution persistence
const state = $getWorkflowStaticData('global');
const key = 'lastMsg_' + chatId;

if (state[key] && (now - state[key]) < rateLimitMs) {
  return []; // Skip processing — too fast
}

state[key] = now;
return [$input.first()];

API Error Detection

Add an IF node after the Claude API call to check for errors:

Condition: {{ $json.error ? true : false }}
True branch: Send an apologetic fallback message to the user
False branch: Continue to the response formatting and Telegram send node

Input Validation

Add another Code node before the API call to filter out non-text messages:

const message = $json.message;

// Skip if no text content (e.g., stickers, images, etc.)
if (!message.text || message.text.trim() === '') {
  return [];
}

// Skip bot commands that should be handled separately
if (message.text.startsWith('/clear') || message.text.startsWith('/help')) {
  return [];
}

return [$input.first()];

Step 7: Deployment and Production Tips

Use a VPS for Self-Hosting

A $6/month DigitalOcean droplet (or equivalent) is ideal. Here's a production-ready Docker Compose configuration:

version: '3.8'

services:
  n8n:
    image: n8nio/n8n
    restart: always
    ports:
      - "5678:5678"
    environment:
      - N8N_HOST=your-domain.com
      - N8N_PORT=5678
      - N8N_PROTOCOL=https
      - WEBHOOK_URL=https://your-domain.com/
      - NODE_ENV=production
      - N8N_ENCRYPTION_KEY=your-random-encryption-key
      - ANTHROPIC_API_KEY=sk-ant-xxxxx
    volumes:
      - n8n_data:/home/node/.n8n

  caddy:
    image: caddy:2
    restart: always
    ports:
      - "80:80"
      - "443:443"
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile
      - caddy_data:/data

volumes:
  n8n_data:
  caddy_data:

Caddyfile for automatic HTTPS:

your-domain.com {
    reverse_proxy n8n:5678
}

Deploy with:

docker compose up -d

Caddy automatically provisions and renews SSL certificates via Let's Encrypt. No manual HTTPS configuration needed.

Monitoring

Create a separate n8n workflow with a Schedule Trigger (every 5 minutes) that pings your bot and alerts you if it's down. Route the alert branch to your personal Telegram, email, or Slack.

Cost Optimization Strategies

Strategy	Estimated Savings
Cache frequent questions with a Redis or Postgres lookup	30-50% fewer API calls
Route simple queries to `claude-3-haiku`, complex ones to `claude-sonnet-4-20250514`	~60% cost reduction
Set `max_tokens` based on expected response length	Prevents paying for unused token capacity
Add a "typing..." indicator while Claude processes	Improves UX, reduces duplicate messages

Complete Workflow JSON (Import-Ready)

Here's the full n8n workflow you can import directly into your instance:

{
  "name": "Claude Telegram Bot",
  "nodes": [
    {
      "parameters": {
        "updates": ["message"]
      },
      "name": "Telegram Trigger",
      "type": "n8n-nodes-base.telegramTrigger",
      "typeVersion": 1.1,
      "position": [250, 300],
      "credentials": {
        "telegramApi": {
          "id": "1",
          "name": "Telegram Bot"
        }
      }
    },
    {
      "parameters": {
        "method": "POST",
        "url": "https://api.anthropic.com/v1/messages",
        "authentication": "genericCredentialType",
        "genericAuthType": "httpHeaderAuth",
        "sendHeaders": true,
        "headerParameters": {
          "parameters": [
            { "name": "x-api-key", "value": "={{ $env.ANTHROPIC_API_KEY }}" },
            { "name": "anthropic-version", "value": "2023-06-01" },
            { "name": "content-type", "value": "application/json" }
          ]
        },
        "sendBody": true,
        "specifyBody": "json",
        "jsonBody": "={{ JSON.stringify({ model: 'claude-sonnet-4-20250514', max_tokens: 1024, system: 'You are a helpful AI assistant on Telegram. Keep responses concise and formatted for mobile reading.', messages: [{ role: 'user', content: $json.message.text }] }) }}"
      },
      "name": "Claude API",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4.2,
      "position": [500, 300],
      "credentials": {
        "httpHeaderAuth": {
          "id": "2",
          "name": "Anthropic API"
        }
      }
    },
    {
      "parameters": {
        "chatId": "={{ $('Telegram Trigger').item.json.message.chat.id }}",
        "text": "={{ $json.content[0].text }}",
        "additionalFields": {
          "parse_mode": "Markdown"
        }
      },
      "name": "Send Response",
      "type": "n8n-nodes-base.telegram",
      "typeVersion": 1.2,
      "position": [750, 300],
      "credentials": {
        "telegramApi": {
          "id": "1",
          "name": "Telegram Bot"
        }
      }
    }
  ],
  "connections": {
    "Telegram Trigger": {
      "main": [[{ "node": "Claude API", "type": "main", "index": 0 }]]
    },
    "Claude API": {
      "main": [[{ "node": "Send Response", "type": "main", "index": 0 }]]
    }
  }
}

To import: In n8n, go to Workflows > Import from File/URL and paste this JSON.

What's Next

Once your basic bot is running, here are the most impactful enhancements:

Conversation memory — Store chat history in Postgres or Redis, and send previous messages as context to Claude for multi-turn conversations
Tool use — Leverage Claude's function calling to let your bot search the web, check weather, query databases, or call external APIs
Multi-modal input — Handle images sent to the bot using Claude's vision capabilities
User authentication — Restrict access to authorized Telegram user IDs using an allowlist
Analytics dashboard — Log all interactions to Google Sheets or a database for usage insights

I'm building all of these into OpenClaw. Follow me for the next article in this series.

Wrapping Up

The Claude API + Telegram + n8n stack is one of the fastest paths from idea to production AI bot. You get Claude's reasoning quality, Telegram's massive user base, and n8n's visual workflow builder — all without writing a traditional backend.

Total cost: ~$6/month for a VPS plus a few dollars in API usage. Compare that to enterprise chatbot platforms charging $200+/month.

The complete workflow JSON above is ready to import. Add your API keys, and you'll have a working AI bot in under 30 minutes.

If this was helpful, drop a reaction and follow me for Part 2 where we add conversation memory and Claude's tool-use capabilities.

Ryosuke Matsumiya — AI Automation Engineer. Building OpenClaw, an AI-powered Telegram agent. Passionate about n8n, Claude API, and workflow automation.

5 n8n Automation Workflows That Save Me 10+ Hours Every Week

松宮亮介 — Mon, 30 Mar 2026 15:23:49 +0000

I replaced Zapier, three Python scripts, and a VA with five n8n workflows. Here's exactly how each one works, with importable JSON.

Why n8n Over Zapier or Make?

Before diving into the workflows, let me address the obvious question: why n8n?

I've used Zapier (3 years), Make/Integromat (1 year), and n8n (2 years). Here's my honest comparison:

Feature	Zapier	Make	n8n
Pricing	$29+/mo for 750 tasks	$9+/mo for 10K ops	Free (self-hosted)
Self-hosting	No	No	Yes
Code nodes	Limited	JavaScript only	JavaScript + Python
AI integrations	Basic	Basic	Native Claude, GPT, etc.
Complex logic	Painful	Better	Full programming power
Data privacy	Cloud only	Cloud only	Your server, your data

The dealbreaker for me was cost. Running 50,000+ operations/month on Zapier would cost $100+. On n8n self-hosted, it costs the $6/month VPS fee — regardless of volume.

Let's look at the five workflows that save me the most time.

Workflow 1: Automated Lead Management Pipeline

Time saved: ~3 hours/week

When a new lead fills out my contact form (Google Forms), this workflow automatically:

Adds the lead to a Google Sheet with a timestamp
Sends a personalized welcome email
Posts a notification to my team's Slack channel
Creates a follow-up reminder for 3 days later

How It Works

Key Nodes

{
  "name": "Lead Pipeline",
  "nodes": [
    {
      "parameters": {
        "formId": "YOUR_FORM_ID"
      },
      "name": "New Form Submission",
      "type": "n8n-nodes-base.googleFormsTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "operation": "append",
        "sheetId": "YOUR_SHEET_ID",
        "range": "Leads!A:F",
        "columns": "timestamp,name,email,company,source,status"
      },
      "name": "Add to Sheet",
      "type": "n8n-nodes-base.googleSheets",
      "typeVersion": 4,
      "position": [450, 300]
    },
    {
      "parameters": {
        "resource": "message",
        "operation": "send",
        "to": "={{ $json.email }}",
        "subject": "Thanks for reaching out, {{ $json.name }}!",
        "emailType": "html",
        "message": "<h2>Hi {{ $json.name }},</h2><p>Thanks for your interest. I'll review your request and get back to you within 24 hours.</p>"
      },
      "name": "Welcome Email",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 2,
      "position": [650, 200]
    },
    {
      "parameters": {
        "channel": "#leads",
        "text": "New lead from {{ $json.name }} ({{ $json.company }})"
      },
      "name": "Slack Alert",
      "type": "n8n-nodes-base.slack",
      "typeVersion": 2,
      "position": [650, 400]
    }
  ]
}

Pro Tip: Lead Scoring

Add a Code node after the Google Sheets node to score leads automatically:

const lead = .first().json;
let score = 0;

// Score based on company size indicators
if (lead.company && lead.company.length > 0) score += 20;
if (lead.source === 'referral') score += 30;
if (lead.source === 'google') score += 10;

// Add score to the data
return [{ json: { ...lead, leadScore: score, priority: score >= 30 ? 'high' : 'normal' } }];

Workflow 2: Social Media Auto-Scheduling

Time saved: ~2 hours/week

I write content in a Google Doc once a week. This workflow takes that content and schedules posts across multiple platforms.

How It Works

Key Nodes

{
  "name": "Social Media Scheduler",
  "nodes": [
    {
      "parameters": {
        "rule": {
          "interval": [{ "triggerAtHour": 9, "triggerAtDay": 1 }]
        }
      },
      "name": "Monday 9AM",
      "type": "n8n-nodes-base.scheduleTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "operation": "read",
        "sheetId": "YOUR_SHEET_ID",
        "range": "Posts!A:E"
      },
      "name": "Get Pending Posts",
      "type": "n8n-nodes-base.googleSheets",
      "typeVersion": 4,
      "position": [450, 300]
    },
    {
      "parameters": {
        "conditions": {
          "string": [
            {
              "value1": "={{ $json.platform }}",
              "value2": "twitter"
            }
          ]
        }
      },
      "name": "Route by Platform",
      "type": "n8n-nodes-base.switch",
      "typeVersion": 3,
      "position": [650, 300]
    }
  ]
}

Platform-Specific Posting

For Twitter/X, use the HTTP Request node with the Twitter API v2:

// Code node to format for Twitter (280 char limit)
const post = .first().json;
let tweet = post.content;

if (tweet.length > 280) {
  tweet = tweet.substring(0, 277) + '...';
}

return [{ json: { text: tweet, originalId: post.id } }];

For LinkedIn, use the HTTP Request node with the LinkedIn Share API. The key is formatting the post body correctly:

{
  "author": "urn:li:person:YOUR_PERSON_ID",
  "lifecycleState": "PUBLISHED",
  "specificContent": {
    "com.linkedin.ugc.ShareContent": {
      "shareCommentary": { "text": "{{ $json.content }}" },
      "shareMediaCategory": "NONE"
    }
  },
  "visibility": {
    "com.linkedin.ugc.MemberNetworkVisibility": "PUBLIC"
  }
}

Workflow 3: AI-Powered Email Draft Generator

Time saved: ~2.5 hours/week

This is my favorite workflow. It monitors my Gmail for emails that need replies, uses Claude API to draft responses, and saves the drafts for my review.

How It Works

Key Nodes

{
  "name": "AI Email Drafter",
  "nodes": [
    {
      "parameters": {
        "pollTimes": { "item": [{ "mode": "everyMinute", "minute": 5 }] },
        "filters": { "labelIds": ["Label_needs_reply"] }
      },
      "name": "New Email",
      "type": "n8n-nodes-base.gmailTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "method": "POST",
        "url": "https://api.anthropic.com/v1/messages",
        "sendHeaders": true,
        "headerParameters": {
          "parameters": [
            { "name": "x-api-key", "value": "={{ $env.ANTHROPIC_API_KEY }}" },
            { "name": "anthropic-version", "value": "2023-06-01" },
            { "name": "content-type", "value": "application/json" }
          ]
        },
        "sendBody": true,
        "specifyBody": "json"
      },
      "name": "Claude Draft",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4.2,
      "position": [500, 300]
    },
    {
      "parameters": {
        "resource": "draft",
        "operation": "create",
        "subject": "Re: {{ $('New Email').item.json.subject }}",
        "message": "={{ $json.content[0].text }}",
        "options": {
          "threadId": "={{ $('New Email').item.json.threadId }}"
        }
      },
      "name": "Save Draft",
      "type": "n8n-nodes-base.gmail",
      "typeVersion": 2,
      "position": [750, 300]
    }
  ]
}

Why Drafts, Not Auto-Send?

I intentionally save as drafts rather than auto-sending. AI-generated emails still need a human eye for:

Tone — Claude is good but not perfect at matching your voice
Accuracy — Factual claims should be verified
Relationships — Some emails need a more personal touch

The time savings come from having a 90%-ready draft instead of starting from a blank page. I typically spend 10-20 seconds reviewing each draft versus 3-5 minutes writing from scratch.

Workflow 4: Web Scraping + Data Processing + Report Generation

Time saved: ~2 hours/week

Every Friday, this workflow scrapes competitor pricing pages, processes the data, generates a comparison report, and emails it to me.

How It Works

Key Nodes

{
  "name": "Competitor Price Monitor",
  "nodes": [
    {
      "parameters": {
        "rule": {
          "interval": [{ "triggerAtHour": 6, "triggerAtDay": 5 }]
        }
      },
      "name": "Friday 6AM",
      "type": "n8n-nodes-base.scheduleTrigger",
      "typeVersion": 1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "jsCode": "const urls = [
  { name: 'Competitor A', url: 'https://competitor-a.com/pricing' },
  { name: 'Competitor B', url: 'https://competitor-b.com/pricing' }
];
return urls.map(u => ({ json: u }));"
      },
      "name": "URL List",
      "type": "n8n-nodes-base.code",
      "typeVersion": 2,
      "position": [450, 300]
    },
    {
      "parameters": {
        "url": "={{ $json.url }}",
        "options": { "timeout": 10000 }
      },
      "name": "Fetch Page",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4.2,
      "position": [650, 300]
    }
  ]
}

Data Extraction Code Node

const html = .first().json.data;
const name = .first().json.name;

// Extract pricing data (customize regex for each site)
const priceRegex = /$(d+(?:.d{2})?)s*(?:/s*mo|pers*month)/gi;
const prices = [];
let match;

while ((match = priceRegex.exec(html)) !== null) {
  prices.push(parseFloat(match[1]));
}

return [{
  json: {
    competitor: name,
    prices: prices,
    scrapedAt: new Date().toISOString()
  }
}];

Report Generation with Claude

After collecting all competitor data, merge the results and send them to Claude for analysis:

const allData = .all().map(item => item.json);

const prompt = "Analyze this competitor pricing data and write a brief report 
highlighting key changes, trends, and actionable insights:

"
  + JSON.stringify(allData, null, 2);

return [{ json: { prompt } }];

Workflow 5: Telegram Bot for Task Management

Time saved: ~1.5 hours/week

This workflow turns Telegram into a lightweight task manager. I send tasks via Telegram, and they are automatically organized, prioritized, and synced with a Google Sheet.

Commands

Command	Action
`/add Buy groceries`	Adds a new task
`/list`	Shows all pending tasks
`/done 3`	Marks task #3 as complete
`/priority 2 high`	Sets task #2 to high priority

How It Works

Key Nodes

{
  "name": "Telegram Task Manager",
  "nodes": [
    {
      "parameters": {
        "updates": ["message"]
      },
      "name": "Telegram Trigger",
      "type": "n8n-nodes-base.telegramTrigger",
      "typeVersion": 1.1,
      "position": [250, 300]
    },
    {
      "parameters": {
        "rules": {
          "rules": [
            { "value": "/add", "output": 0 },
            { "value": "/list", "output": 1 },
            { "value": "/done", "output": 2 },
            { "value": "/priority", "output": 3 }
          ]
        },
        "dataType": "string",
        "value1": "={{ $json.message.text.split(' ')[0] }}"
      },
      "name": "Command Router",
      "type": "n8n-nodes-base.switch",
      "typeVersion": 3,
      "position": [450, 300]
    }
  ]
}

Add Task Handler

const text = .message.text;
const taskText = text.replace(/^/adds+/, '');
const chatId = .message.chat.id;

if (!taskText) {
  return [{ json: { chatId, response: 'Usage: /add <task description>' } }];
}

const task = {
  id: Date.now(),
  text: taskText,
  status: 'pending',
  priority: 'normal',
  createdAt: new Date().toISOString(),
  chatId: chatId
};

return [{ json: { ...task, response: 'Task added: ' + taskText } }];

List Tasks Handler

const tasks = .all().map(item => item.json);
const pending = tasks.filter(t => t.status === 'pending');

if (pending.length === 0) {
  return [{ json: { response: 'No pending tasks. Use /add to create one.' } }];
}

const list = pending.map((t, i) => {
  const pri = t.priority === 'high' ? '[!]' : '   ';
  return (i + 1) + '. ' + pri + ' ' + t.text;
}).join('
');

return [{ json: { response: 'Pending tasks:

' + list } }];

Performance Tips for n8n Workflows

After running these workflows for over a year, here are my top optimization tips:

1. Use Static Data for Caching

n8n's () persists data between executions without needing a database:

const cache = ('global');
const cacheKey = 'lastResult_' + someId;
const cacheMaxAge = 3600000; // 1 hour

if (cache[cacheKey] && (Date.now() - cache[cacheKey].timestamp) < cacheMaxAge) {
  return [{ json: cache[cacheKey].data }];
}

// ... fetch fresh data ...
cache[cacheKey] = { data: freshData, timestamp: Date.now() };

2. Error Handling on Every API Call

Always add an error handler. API calls fail more often than you think — rate limits, network issues, expired tokens:

try {
  const result = .first().json;
  if (result.error) throw new Error(result.error.message);
  return [{ json: result }];
} catch (err) {
  return [{ json: { error: true, message: err.message, fallback: 'Default response' } }];
}

3. Batch Operations Where Possible

Instead of making 100 individual API calls, batch them:

const items = .all().map(item => item.json);
const batchSize = 10;
const batches = [];

for (let i = 0; i < items.length; i += batchSize) {
  batches.push(items.slice(i, i + batchSize));
}

return batches.map(batch => ({ json: { batch } }));

4. Monitor Execution Times

n8n shows execution time per node. If a workflow takes more than 30 seconds, look for:

Unnecessary API calls that could be cached
Sequential operations that could run in parallel (use the Split In Batches node)
Large payloads being passed between nodes (extract only what you need)

5. Use Environment Variables

Never hardcode API keys or sensitive data in workflows:

// In n8n, reference environment variables like this:
const apiKey = .ANTHROPIC_API_KEY;
const webhookUrl = .SLACK_WEBHOOK_URL;

Set them in your Docker Compose or .env file, not in the workflow JSON.

Wrapping Up

These five workflows have fundamentally changed how I work. The initial setup takes a few hours each, but the compounding time savings are enormous:

Workflow	Weekly Time Saved
Lead Management Pipeline	3 hours
Social Media Scheduling	2 hours
AI Email Draft Generator	2.5 hours
Web Scraping + Reports	2 hours
Telegram Task Manager	1.5 hours
Total	11 hours/week

That is over 500 hours per year redirected from repetitive tasks to actual creative and strategic work.

The best part? Every one of these workflows runs on a single $6/month VPS. No SaaS subscriptions, no per-execution fees, no vendor lock-in.

Start with one workflow — whichever solves your biggest pain point — and build from there. The JSON snippets above are ready to import into n8n and customize for your needs.

If you want to see how I combine all of these with Claude API for even more automation, check out my previous article: Building an AI-Powered Telegram Bot with Claude API and n8n.

Ryosuke Matsumiya — AI Automation Engineer. Building OpenClaw, an AI-powered Telegram agent. Passionate about n8n, Claude API, and workflow automation.

Building an AI-Powered Telegram Bot with Claude API and n8n: A Complete Guide

松宮亮介 — Mon, 30 Mar 2026 15:23:30 +0000

Build a production-ready AI chatbot on Telegram using Anthropic's Claude API and n8n in under 30 minutes, with zero traditional backend code.

Why Claude API + Telegram + n8n?

There's a better way. By combining three tools — Claude API for intelligence, Telegram for the user interface, and n8n for orchestration — you get a production-grade chatbot with:

Claude 3.5 Sonnet / Claude 4 as the reasoning engine — far more nuanced than GPT for following complex instructions
Telegram as a zero-friction UI that 900M+ users already have installed
n8n as the glue layer — visual workflow automation that you self-host, so there are no per-execution fees like Zapier

I've been running this exact stack for OpenClaw, an AI-powered Telegram agent, and it handles thousands of messages daily without a single line of Express.js. Let me show you how to build it.

Prerequisites

Before we start, make sure you have:

Requirement	Details
Node.js	v18+ (required for n8n)
n8n	Self-hosted or n8n Cloud account
Anthropic API Key	Get one at console.anthropic.com
Telegram Account	To create a bot via BotFather

Estimated cost: ~$0.003 per message with Claude 3.5 Sonnet. For most personal-use bots, you'll spend less than $5/month.

Step 1: Create Your Telegram Bot with BotFather

Open Telegram and search for @botfather. This is Telegram's official bot for creating bots.

/start
/newbot

BotFather will ask two questions:

Bot name: The display name (e.g., "My Claude Assistant")
Bot username: Must end in bot (e.g., my_claude_assistant_bot)

You'll receive a Bot Token that looks like this:

7123456789:AAH_your-token-string-here

Save this token — you'll need it in Step 3.

Optional but Recommended: Configure Bot Settings

/setdescription - Add a description for your bot
/setabouttext   - Set the "About" section
/setuserpic     - Upload a profile picture
/setcommands    - Define command menu

Here's a useful command list to set via /setcommands:

start - Start the bot
help - Show available commands
clear - Reset conversation context

Step 2: Install and Set Up n8n

Option A: Self-Hosted (Recommended for Production)

# Install n8n globally
npm install -g n8n

# Or use Docker (preferred for production)
docker run -d \
  --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  -e WEBHOOK_URL=https://your-domain.com/ \
  n8nio/n8n

Important: For Telegram webhooks to work, n8n must be accessible via HTTPS on a public URL. If you're developing locally, use a tunnel service like ngrok or cloudflared.

# Quick tunnel for local development
npx ngrok http 5678

Option B: n8n Cloud

Sign up at app.n8n.cloud. The free tier gives you 2,500 executions/month — more than enough for testing. Webhooks work out of the box, no tunnel required.

Verify Installation

Open http://localhost:5678 (or your cloud URL). You should see the n8n canvas editor. Create a new workflow and name it "Claude Telegram Bot".

Step 3: Configure the Telegram Trigger Node

In your new workflow, click "Add first step" and search for Telegram.

Select "Telegram Trigger" — this node listens for incoming messages to your bot.

Configure Credentials

Click "Create New Credential"
Paste your Bot Token from Step 1
Click "Save"

Trigger Settings

Setting	Value
Updates	`message`
Additional Fields	Leave default

The node configuration JSON:

{
  "parameters": {
    "updates": ["message"]
  },
  "name": "Telegram Trigger",
  "type": "n8n-nodes-base.telegramTrigger",
  "typeVersion": 1.1,
  "position": [250, 300],
  "credentials": {
    "telegramApi": {
      "id": "YOUR_CREDENTIAL_ID",
      "name": "Telegram Bot"
    }
  }
}

Test the Trigger

Click "Listen for test event" in n8n
Send any message to your bot on Telegram
You should see the incoming message data appear in n8n

The payload includes message.text, message.chat.id, message.from.first_name, and more. We'll use these fields in subsequent steps.

Step 4: Connect the Claude API (Anthropic Node)

Add a new node after the Telegram Trigger. Search for "Anthropic" or "HTTP Request".

Option A: Using the Built-in Anthropic Node (n8n v1.30+)

n8n v1.30+ ships with a native Anthropic node. Configure it as follows:

Credential: Add your Anthropic API key
Model: claude-sonnet-4-20250514 (best balance of speed, quality, and cost)
Prompt: Use an expression to inject the user's Telegram message:

{{ $json.message.text }}

Option B: Using HTTP Request Node (Any n8n Version)

If your n8n version doesn't have the native Anthropic node, the HTTP Request node works perfectly:

{
  "parameters": {
    "method": "POST",
    "url": "https://api.anthropic.com/v1/messages",
    "authentication": "genericCredentialType",
    "genericAuthType": "httpHeaderAuth",
    "sendHeaders": true,
    "headerParameters": {
      "parameters": [
        {
          "name": "x-api-key",
          "value": "YOUR_ANTHROPIC_API_KEY"
        },
        {
          "name": "anthropic-version",
          "value": "2023-06-01"
        },
        {
          "name": "content-type",
          "value": "application/json"
        }
      ]
    },
    "sendBody": true,
    "specifyBody": "json",
    "jsonBody": "={{ JSON.stringify({ model: 'claude-sonnet-4-20250514', max_tokens: 1024, system: 'You are a helpful AI assistant on Telegram. Keep responses concise and formatted for mobile reading. Use short paragraphs and bullet points when appropriate.', messages: [{ role: 'user', content: $json.message.text }] }) }}"
  },
  "name": "Claude API",
  "type": "n8n-nodes-base.httpRequest",
  "typeVersion": 4.2,
  "position": [500, 300]
}

System Prompt Best Practices for Telegram Bots

The system prompt determines your bot's personality and behavior. Here's one that works well:

You are a helpful AI assistant on Telegram. Follow these rules:
1. Keep responses under 4096 characters (Telegram's message limit).
2. Use short paragraphs — mobile screens are small.
3. Use bullet points and numbered lists for structured information.
4. If a question is ambiguous, give the most likely answer first,
   then briefly mention alternatives.
5. Format code blocks with triple backticks and language identifiers.

Step 5: Send the Response Back to Telegram

Add a "Telegram" node (the action node, not the Trigger) after the Claude API node.

Configuration

Setting	Value
Resource	Send Message
Chat ID	`{{ $('Telegram Trigger').item.json.message.chat.id }}`
Text	Depends on Step 4 method (see below)
Parse Mode	Markdown

For the Text field:

If using the native Anthropic node:

{{ $json.message.content[0].text }}

If using the HTTP Request node:

{{ $json.content[0].text }}

Handle Long Responses

Telegram enforces a 4096-character limit per message. Add a Code node between Claude API and the Telegram send node to split long responses:

const text = $input.first().json.content[0].text;
const chatId = $('Telegram Trigger').first().json.message.chat.id;
const maxLength = 4000; // Leave buffer for Markdown formatting

if (text.length <= maxLength) {
  return [{ json: { chatId, text } }];
}

// Split into chunks at natural paragraph boundaries
const chunks = [];
let remaining = text;

while (remaining.length > 0) {
  if (remaining.length <= maxLength) {
    chunks.push(remaining);
    break;
  }

  let splitIndex = remaining.lastIndexOf('\n\n', maxLength);
  if (splitIndex === -1) splitIndex = remaining.lastIndexOf('\n', maxLength);
  if (splitIndex === -1) splitIndex = remaining.lastIndexOf(' ', maxLength);
  if (splitIndex === -1) splitIndex = maxLength;

  chunks.push(remaining.substring(0, splitIndex));
  remaining = remaining.substring(splitIndex).trimStart();
}

return chunks.map(chunk => ({ json: { chatId, text: chunk } }));

Then update the Telegram send node to use {{ $json.text }} for the text and {{ $json.chatId }} for the chat ID.

Step 6: Error Handling and Fallback

Production bots need graceful error handling. Here's how to make yours resilient.

Add an Error Trigger Workflow

Create a separate workflow called "Bot Error Handler":

Add an Error Trigger node
Connect it to a Telegram Send Message node
Configure the error message:

Sorry, I encountered an error processing your request.
Please try again in a moment.

If this persists, use /clear to reset the conversation.

Rate Limiting

Claude API has rate limits, and users sometimes spam messages. Add a Code node before the API call:

const chatId = $json.message.chat.id;
const now = Date.now();
const rateLimitMs = 2000; // Minimum 2 seconds between messages

// Use n8n's static data for cross-execution persistence
const state = $getWorkflowStaticData('global');
const key = 'lastMsg_' + chatId;

if (state[key] && (now - state[key]) < rateLimitMs) {
  return []; // Skip processing — too fast
}

state[key] = now;
return [$input.first()];

API Error Detection

Add an IF node after the Claude API call to check for errors:

Condition: {{ $json.error ? true : false }}
True branch: Send an apologetic fallback message to the user
False branch: Continue to the response formatting and Telegram send node

Input Validation

Add another Code node before the API call to filter out non-text messages:

const message = $json.message;

// Skip if no text content (e.g., stickers, images, etc.)
if (!message.text || message.text.trim() === '') {
  return [];
}

// Skip bot commands that should be handled separately
if (message.text.startsWith('/clear') || message.text.startsWith('/help')) {
  return [];
}

return [$input.first()];

Step 7: Deployment and Production Tips

Use a VPS for Self-Hosting

A $6/month DigitalOcean droplet (or equivalent) is ideal. Here's a production-ready Docker Compose configuration:

version: '3.8'

services:
  n8n:
    image: n8nio/n8n
    restart: always
    ports:
      - "5678:5678"
    environment:
      - N8N_HOST=your-domain.com
      - N8N_PORT=5678
      - N8N_PROTOCOL=https
      - WEBHOOK_URL=https://your-domain.com/
      - NODE_ENV=production
      - N8N_ENCRYPTION_KEY=your-random-encryption-key
      - ANTHROPIC_API_KEY=sk-ant-xxxxx
    volumes:
      - n8n_data:/home/node/.n8n

  caddy:
    image: caddy:2
    restart: always
    ports:
      - "80:80"
      - "443:443"
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile
      - caddy_data:/data

volumes:
  n8n_data:
  caddy_data:

Caddyfile for automatic HTTPS:

your-domain.com {
    reverse_proxy n8n:5678
}

Deploy with:

docker compose up -d

Caddy automatically provisions and renews SSL certificates via Let's Encrypt. No manual HTTPS configuration needed.

Monitoring

Create a separate n8n workflow with a Schedule Trigger (every 5 minutes) that pings your bot and alerts you if it's down. Route the alert branch to your personal Telegram, email, or Slack.

Cost Optimization Strategies

Strategy	Estimated Savings
Cache frequent questions with a Redis or Postgres lookup	30-50% fewer API calls
Route simple queries to `claude-3-haiku`, complex ones to `claude-sonnet-4-20250514`	~60% cost reduction
Set `max_tokens` based on expected response length	Prevents paying for unused token capacity
Add a "typing..." indicator while Claude processes	Improves UX, reduces duplicate messages

Complete Workflow JSON (Import-Ready)

Here's the full n8n workflow you can import directly into your instance:

{
  "name": "Claude Telegram Bot",
  "nodes": [
    {
      "parameters": {
        "updates": ["message"]
      },
      "name": "Telegram Trigger",
      "type": "n8n-nodes-base.telegramTrigger",
      "typeVersion": 1.1,
      "position": [250, 300],
      "credentials": {
        "telegramApi": {
          "id": "1",
          "name": "Telegram Bot"
        }
      }
    },
    {
      "parameters": {
        "method": "POST",
        "url": "https://api.anthropic.com/v1/messages",
        "authentication": "genericCredentialType",
        "genericAuthType": "httpHeaderAuth",
        "sendHeaders": true,
        "headerParameters": {
          "parameters": [
            { "name": "x-api-key", "value": "={{ $env.ANTHROPIC_API_KEY }}" },
            { "name": "anthropic-version", "value": "2023-06-01" },
            { "name": "content-type", "value": "application/json" }
          ]
        },
        "sendBody": true,
        "specifyBody": "json",
        "jsonBody": "={{ JSON.stringify({ model: 'claude-sonnet-4-20250514', max_tokens: 1024, system: 'You are a helpful AI assistant on Telegram. Keep responses concise and formatted for mobile reading.', messages: [{ role: 'user', content: $json.message.text }] }) }}"
      },
      "name": "Claude API",
      "type": "n8n-nodes-base.httpRequest",
      "typeVersion": 4.2,
      "position": [500, 300],
      "credentials": {
        "httpHeaderAuth": {
          "id": "2",
          "name": "Anthropic API"
        }
      }
    },
    {
      "parameters": {
        "chatId": "={{ $('Telegram Trigger').item.json.message.chat.id }}",
        "text": "={{ $json.content[0].text }}",
        "additionalFields": {
          "parse_mode": "Markdown"
        }
      },
      "name": "Send Response",
      "type": "n8n-nodes-base.telegram",
      "typeVersion": 1.2,
      "position": [750, 300],
      "credentials": {
        "telegramApi": {
          "id": "1",
          "name": "Telegram Bot"
        }
      }
    }
  ],
  "connections": {
    "Telegram Trigger": {
      "main": [[{ "node": "Claude API", "type": "main", "index": 0 }]]
    },
    "Claude API": {
      "main": [[{ "node": "Send Response", "type": "main", "index": 0 }]]
    }
  }
}

To import: In n8n, go to Workflows > Import from File/URL and paste this JSON.

What's Next

Once your basic bot is running, here are the most impactful enhancements:

Conversation memory — Store chat history in Postgres or Redis, and send previous messages as context to Claude for multi-turn conversations
Tool use — Leverage Claude's function calling to let your bot search the web, check weather, query databases, or call external APIs
Multi-modal input — Handle images sent to the bot using Claude's vision capabilities
User authentication — Restrict access to authorized Telegram user IDs using an allowlist
Analytics dashboard — Log all interactions to Google Sheets or a database for usage insights

I'm building all of these into OpenClaw. Follow me for the next article in this series.

Wrapping Up

Total cost: ~$6/month for a VPS plus a few dollars in API usage. Compare that to enterprise chatbot platforms charging $200+/month.

The complete workflow JSON above is ready to import. Add your API keys, and you'll have a working AI bot in under 30 minutes.

If this was helpful, drop a reaction and follow me for Part 2 where we add conversation memory and Claude's tool-use capabilities.

Ryosuke Matsumiya — AI Automation Engineer. Building OpenClaw, an AI-powered Telegram agent. Passionate about n8n, Claude API, and workflow automation.