Claude Workflows: Chain Prompts for Complex Automation

What Is a Claude Workflow?

A Claude workflow is any automated process where Claude is one step in a larger chain. Instead of manually prompting Claude and copying output, workflows:

1. Trigger automatically (schedule, webhook, event)
2. Gather context from external sources (CRM, database, email)
3. Send context + task to Claude API
4. Process Claude's response (save to database, send email, update CRM)
5. Continue the chain (notify team, trigger next step)

Claude becomes an intelligent processing node inside your business operations.

Core Workflow Patterns

Enrichment Pattern: Input data (lead, document, record) → Claude adds analysis, classification, or insights → enriched output stored/sent.

Example: New HubSpot contact → Claude researches company context + generates personalised talking points → saved to contact record.

Transform Pattern: Raw content → Claude restructures, reformats, or converts → structured output.

Example: Customer support email → Claude → structured ticket with category, priority, and draft response → Zendesk.

Generate Pattern: Brief or parameters → Claude creates content → content stored or published.

Example: Scheduled trigger + keyword list → Claude → SEO article → CMS draft.

Analyse and Route Pattern: Input → Claude classifies/analyses → routes to appropriate next step.

Example: New support ticket → Claude classifies intent and urgency → route to tier 1, tier 2, or escalation queue.

Monitor and Alert Pattern: Regular trigger → Claude reviews state of something → alert if notable.

Example: Daily trigger → scrape competitor pages → Claude compares to baseline → Slack alert if significant change detected.

Building Workflows with n8n

n8n is the best tool for building Claude workflows if you want flexibility and self-hosting capability.

Basic Claude API call in n8n:

1. Add an HTTP Request node
2. Method: POST
3. URL: https://api.anthropic.com/v1/messages
4. Headers: anthropic-version: 2023-06-01, x-api-key: {{$env.ANTHROPIC_API_KEY}}, content-type: application/json
5. Body:

This is the building block. Everything else is connecting nodes around this core call.

Real workflow — Lead enrichment:

• Trigger: New HubSpot contact created
• HTTP: Get company info via Clearbit API
• HTTP: Claude API — "Given this person's role and company, what are the most relevant use cases for our product and likely objections?"
• HubSpot: Update contact with AI-generated talking points

Building Workflows with Make.com

Make.com is better for non-technical users. Use the HTTP module for Claude API calls:

1. Add HTTP module → Make a request
2. URL: https://api.anthropic.com/v1/messages
3. Method: POST
4. Headers: add anthropic-version, x-api-key, content-type
5. Body type: Raw, JSON format

Connect Make's native modules (Gmail, HubSpot, Airtable, Slack) around the Claude HTTP call.

Workflow Examples

Daily Sales Digest: Trigger: Every morning at 8am → HubSpot: pull all deals updated in last 24h → Claude: "Summarise the state of our sales pipeline, flag any deals at risk, and suggest 3 priority actions for today" → Email to sales team.

Automated Content Calendar: Trigger: Monday 9am → Airtable: pull this week's topics → Loop: Claude generates article for each topic → CMS: create drafts → Slack: notify content team.

Customer Health Monitoring: Trigger: Daily → Database: pull usage metrics for each account → Claude: classify health (healthy, at-risk, churning) and generate action recommendation → HubSpot: update health score + activity → Alert: notify CSM if at-risk.

New Employee Onboarding: Trigger: New Slack member added → Claude: generate personalised welcome message + first-week guide based on their role → Slack: DM new employee → Calendar: schedule week-1 check-in.

Competitor Intelligence: Trigger: Weekly → Web: scrape competitor blogs and news → Claude: "What's new from our competitors this week? What's significant for us?" → Email report to product and marketing.

Prompt Design for Workflows

Workflow prompts differ from interactive prompts:

Be explicit about output format: Workflows need parseable output. Tell Claude exactly what to produce:

Include all context in the prompt: No back-and-forth in workflows. Everything Claude needs must be in the initial message.

Error handling: Instruct Claude what to return if it can't complete the task properly:

Recommended Tools

• Claude API — Core AI processing node
• n8n — Flexible, self-hostable workflow automation
• Make.com — Visual automation for non-technical teams
• Zapier — Simple workflow automation for common apps
• HubSpot — CRM that connects to Claude workflows
• Airtable — Data layer for content and operations workflows