zapier-workflows

Zapier Workflows Skill

The Problem This Solves

Zapier MCP gives Claude access to 8,000+ individual tools (every Zapier action), but there are critical limitations:

❌ No memory - Claude doesn't remember which tools YOU use or why ❌ No context - Doesn't know when to use specific tools for your workflows ❌ Only one-off actions - Can't trigger your complex, multi-step Zaps ❌ Fresh start every session - All context lost between conversations

The Two Types of Zapier Automation

1. MCP Tools (One-Off Actions)

• Individual Zapier actions (Add row to sheet, Send email, etc.)
• Available via Zapier MCP at https://mcp.zapier.com/mcp/servers
• Great for flexible, on-the-fly automation
• Problem: 8,000+ choices with no guidance on which to use or when

2. Multi-Step Zaps (Webhook-Triggered)

• Complex workflows you've built in Zapier dashboard
• Multiple actions chained together, pre-optimized
• Triggered via webhook URL (POST request)
• Problem: Claude can't trigger these - they're not in the MCP

What This Skill Does

This skill solves both problems by giving Claude persistent memory for your Zapier workflows:

✅ Remembers your MCP tool preferences - "Use Google Sheets for expenses, Notion for tasks" ✅ Knows when/why to use each tool - "Search with Perplexity when researching, not Google" ✅ Triggers multi-step Zaps - "Run my daily digest" = webhook POST to your complex Zap ✅ Self-learning - Claude updates the skill as you teach it, never forgets ✅ Cross-session persistence - Works across all conversations (global install)

What You Get

For Multi-Step Zaps:

• Store webhook URLs and what they do
• Trigger complex workflows just by asking
• Remember when/why to use each Zap
• Document costs, timing, outputs

For MCP Tools:

• Document which tools you prefer for which tasks
• Build reusable workflow patterns
• Store tool-specific preferences (sheet names, formats, etc.)
• Create multi-tool orchestration sequences

Self-Learning:

• Claude automatically updates skill files when you teach it
• Changes persist forever (global install) or per-project (local install)
• No manual editing required - just talk to Claude

Installation & Setup

Installation Location

Global (~/.claude/skills/) - RECOMMENDED:

• Learned patterns persist across ALL projects
• One Zap library for everything
• Preferences carry over to all projects

Project-level (./.claude/skills/):

• Learned patterns ONLY in this project
• Isolated from other projects
• Useful for project-specific workflows

⚠️ Security Warning

IMPORTANT: This skill stores webhook URLs and workflow details in plain text files.

Webhook URLs contain authentication tokens. If someone has your webhook URL, they can trigger your Zaps.

Best practices:

• ✅ Install globally at ~/.claude/skills/ (not in project repos)
• ✅ Add .claude/ to your .gitignore if installed in a project
• ✅ Never commit skill files with real webhook URLs to public repos
• ✅ Regenerate webhook URLs if accidentally exposed
• ✅ Use Zapier's webhook authentication features when available

If you need to share this skill:

• Remove real webhook URLs first
• Replace with placeholder examples
• Or use separate webhook URLs for sharing/testing

Prerequisites

Required:

• Claude Code
• Zapier account (for webhooks and MCP tools)

Optional:

• Additional MCP tools based on your workflows (Perplexity Search, Google Sheets, etc.)

Setting Up Zapier MCP

To connect Zapier's MCP tools to Claude Code:

1. Go to Zapier MCP servers:
2. Visit https://mcp.zapier.com/mcp/servers
3. Login to your Zapier account if prompted

1. Create a new MCP server:
2. Click "New MCP Server" button (top left)
3. In "MCP Client (required)" dropdown, select Claude Code
4. Give your server a name (e.g., "My Zapier Tools")

1. Add tools:
2. Click "Add tools" button
3. Select as many Zapier actions as you want (each becomes an MCP tool)
4. Common tools: Run Zap, Add Row to Google Sheets, Send Email, etc.

1. Connect to Claude Code:
2. Click "Connect" button
3. You'll see a command like this:

``bash claude mcp add zapier https://mcp.zapier.com/api/mcp/mcp -t http -H "Authorization: Bearer ZjFmZGJkN..................1NjBhYzc2MDRlYg==" ``

• Copy and run this command in your terminal

1. Restart Claude Code:
2. Close and reopen Claude Code
3. Your Zapier MCP tools are now available

Tip: You can add more tools later by editing your MCP server in Zapier and running the connect command again.

Creating Webhook-Triggered Zaps

For pre-built, optimized workflows that you want to trigger on-demand:

1. In Zapier dashboard:
2. Create a new Zap
3. Choose "Webhooks by Zapier" as trigger
4. Select "Catch Hook"
5. Copy the webhook URL provided

1. Build your workflow:
2. Add whatever actions you want (API calls, data processing, etc.)
3. Test and optimize the Zap

1. Document it in this skill:
2. Tell Claude about the new Zap (webhook URL, what it does, trigger phrases)
3. Claude will add it to references/zaps.md automatically
4. Now you can trigger it by just asking Claude!

Webhook vs MCP Tools:

• Webhooks: Pre-built, multi-step Zaps you trigger with a POST request. Great for complex, optimized workflows.
• MCP Tools: Individual Zapier actions called directly. Great for flexible, on-the-fly automation.

Self-Improvement Protocol

CRITICAL: This skill can and should edit itself to learn from user feedback.

When the user teaches you something new or corrects your approach:

1. Identify what to update:
2. New Zap to document → Edit references/zaps.md
3. MCP tool preference → Edit references/mcp-patterns.md
4. New workflow pattern → Edit references/mcp-patterns.md

1. Make the edit using Claude Code tools:
2. Read the file first with the Read tool
3. Update with the Edit tool (specify exact old_string and new_string)
4. Confirm the change to the user

1. Update format:

``markdown User: "Use Apollo instead of Clearbit for company data" Claude: [uses Read tool on references/mcp-patterns.md] [uses Edit tool to update the preference] "Updated! I'll use Apollo for company enrichment from now on. This change is now permanent in the skill." ``

What to capture in skill updates:

• ✅ Tool preferences (which tool for which task)
• ✅ Workflow sequences (step-by-step patterns)
• ✅ Error handling approaches
• ✅ Data formatting requirements
• ✅ New Zaps and their details
• ❌ One-off requests (don't clutter the skill)
• ❌ Temporary context (use memory instead)

Decision Logic

When to Use Webhook-Triggered Zaps

Use webhooks when:

• Task is complex, multi-step, and already refined
• User mentions a specific Zap name (check references/zaps.md)
• Deterministic execution is critical
• Task involves 5+ API calls or complex orchestration
• Cost/time efficiency matters (pre-built Zaps are optimized)

When to Use MCP Tool Orchestration

Use MCP tools when:

• Task is simple (1-3 actions)
• Flexibility is needed (parameters change)
• Testing a new workflow pattern
• User explicitly asks to use specific MCP tools

Execution Pattern

1. Listen for triggers: Workflow names, "run", "trigger", "search", "research", etc.
2. Check references: Use Read tool on appropriate reference file for details
3. Check prerequisites:
4. If MCP tools needed but not available: Provide Zapier MCP setup instructions (see below)
5. If MCP tools available but not documented: Trigger tool discovery protocol (see below)
6. If webhook URL needed but not in references: Provide webhook extraction instructions (see below)
7. Execute:
8. For webhook-triggered Zaps: Use Bash tool with curl to POST to webhook URL (webhooks are created in Zapier dashboard wi