wp-product-talk-aug-2025

# 🚀 From Manual to Magical: How We Automated Our WordPress Support Pipeline with AI *A comprehensive guide from WP Product Talk August 2025 with Very Good Plugins* --- <div align="center"> ### 🆕 NEW: Complete Technical Deep-Dive Now Available! <a href="https://wpfusion.com/business/we-built-4-tools-to-fix-our-support-pipeline-and-you-can-have-them-for-free/"> <img src="https://wpfusion.com/wp-content/uploads/2025/08/ai-workflows-aug-2025-1-1024x619.jpg" alt="WP Fusion Blog Post" width="640" height="384"> </a> **[We Built 4 Tools to Fix Our Support Pipeline (and You Can Have Them for Free)](https://wpfusion.com/business/we-built-4-tools-to-fix-our-support-pipeline-and-you-can-have-them-for-free/)** *"We spent about 160 hours in August building these. And yeah, we're giving them all away for free because honestly, we'd love some help maintaining them 😉"* **What you'll get:** Complete implementation guides • Real cost breakdowns • Everything that went hilariously wrong • All 4 tools with installation instructions [**Read the Full Technical Breakdown →**](https://wpfusion.com/business/we-built-4-tools-to-fix-our-support-pipeline-and-you-can-have-them-for-free/) </div> --- <div align="center"> ### 📺 Watch the Full Episode <a href="https://www.youtube.com/watch?v=CVHlPDXGMmY&t=3119s"> <img src="https://i.ytimg.com/vi/CVHlPDXGMmY/maxresdefault.jpg" alt="WP Product Talk - AI Workflows August 2025" width="640" height="360"> </a> **[Watch on YouTube →](https://www.youtube.com/watch?v=CVHlPDXGMmY&t=3119s)** *Deep dive into our complete AI automation setup with live demos, real-world examples, and Q&A session covering everything from Claude Desktop setup to autonomous bug fixing workflows.* **What you'll see:** Live MCP server demos • Claude fixing real bugs • Support workflow automation • Desktop productivity setup • Q&A with the community </div> ## Table of Contents - [The Paradigm Shift](#the-paradigm-shift) - [The Problem We Solved](#the-problem-we-solved) - [The Complete Tool Stack](#the-complete-tool-stack) - [Four Tools That Changed Everything](#four-tools-that-changed-everything) - [The Claude Automation Workflow](#the-claude-automation-workflow) - [Setting Up Your Own AI Pipeline](#setting-up-your-own-ai-pipeline) - [Real Results & ROI](#real-results--roi) - [Lessons Learned & Best Practices](#lessons-learned--best-practices) - [Resources & Next Steps](#resources--next-steps) --- ## 🌟 The Paradigm Shift Something fundamental has changed in how we approach development and support at Very Good Plugins. It's not just about having better tools—it's about a complete mental shift in what's possible. ### The New Reality > "The difficulty now is that everything is fixable with some effort. So little problems I used to ignore now go into my queue. I'm working way longer as a result but getting 100x done." This perfectly captures our experience. Problems that once seemed too tedious or time-consuming to fix are now just a Claude prompt away. That typo in the documentation? Fixed. That edge case bug affecting 3 users? Resolved. That workflow that could be 10% more efficient? Optimized. ### Beyond Code: Financial Freedom One unexpected benefit? AI assistance extends beyond development: > "Even money: I don't worry about it anymore. I can always do a quick trade on some stocks (with LLM assist) and net a few thousand dollars if I need it." When you realize AI can help with everything from code reviews to investment decisions, it fundamentally changes how you approach problems. The question shifts from "Can I do this?" to "How quickly can Claude and I do this?" --- ## 📉 The Problem We Solved ### Before: The Manual Nightmare Our support workflow in July 2024 was, to put it mildly, "extremely manual": 1. **Ticket arrives** → Read everything, try to understand what they're actually asking 2. **It's a bug** → Copy/paste to GitHub, manually format, hope you didn't miss anything 3. **Create worktree** → Terminal dance, remember branch naming convention, pray it works 4. **Fix the thing** → This part's fine, we're developers 5. **Make a PR** → More formatting, link back to ticket, forget something, go back 6. **Write reply** → Start from scratch every time, try to sound friendly but professional 7. **Update FreeScout** → Click around, change status, add tags, close ticket **Result**: A simple bug fix could eat 30+ minutes just in admin overhead. ### The Human Cost When you're doing this 20 times a day: - Your brain turns to mush - You start sending replies with the wrong tone - You promise fixes for things that are actually third-party problems - You forget to link the PR to the ticket - Team morale tanks 😞 We knew something had to change when we were saving $5,700/year by switching to open source tools (FreeScout + FluentCRM) but spending that much in lost productivity every month. --- ## 🛠️ The Complete Tool Stack ### Core AI Infrastructure #### **1. Claude Code - $200/month** The heavyweight champion of our stack. This isn't just about having access to Claude—it's about having **Opus 4.1** with maximum context and capabilities. **Why it's worth every penny:** - Default permissions for new projects (no more permission dance) - Max 20x context window - essential for complex codebases - Server deployment capabilities - run it anywhere - Command-line integration for complex workflows **Pro tip**: Follow [this guide](https://jewelhuq.medium.com/how-to-run-claude-code-on-a-server-and-access-it-anywhere-c31b31488ed9) to run Claude Code on a server and access it from anywhere. #### **2. Cursor IDE - The Developer's Best Friend** While Claude Desktop is amazing for system integration, Cursor is where the magic happens for development: - Best-in-class model selection - Seamless GitHub workflow - Hot-reloadable environment - Built for pair programming with AI **The killer feature**: Cursor + Claude can navigate your entire codebase, understand context across files, and make coherent changes that actually work. #### **3. Claude Desktop + MCP Servers - The Game Changer** > "Can't believe I didn't try it for so long. What a game changer." Claude Desktop isn't just another app—it's a command center that connects to everything: **Desktop Integrations That Feel Like Magic:** - Direct file system access - Browser automation - System control - Application integration **MCP (Model Context Protocol) Servers:** These are the bridges that let Claude talk to your tools. Think of them as APIs that Claude can use natively: - FreeScout (support tickets) - GitHub (code management) - Slack (team communication) - WordPress (content management) - PostgreSQL (database) - And [many more](https://github.com/modelcontextprotocol/servers) **Essential reading**: [6 MCP Servers Every Developer Needs to Try](https://medium.com/coding-nexus/6-mcp-servers-every-developer-needs-to-try-622b3e639403) ### Supporting Cast #### **4. Desktop Productivity Suite** Thanks to Claude Desktop (which set this up autonomously!), our Mac workflow is transformed: - **yabai** - Tiling window manager for macOS - Auto-arrange windows - Keyboard-driven navigation - Virtual desktops on steroids - **skhd** - Simple hotkey daemon - Custom keyboard shortcuts for everything - Launch apps, switch spaces, resize windows - No more clicking around - **sketchybar** - Custom menu bar - System stats at a glance - Quick actions - Clean, minimal interface > "Saving me a ton of time—can't believe how I put up with drag and dropping on OSX for so long" --- ## 🔧 Four Tools That Changed Everything We spent 160 hours in August 2024 building these tools. Here's the complete breakdown: ### Tool #1: WP Clean Admin - The Screenshot Saver **The Problem**: Every documentation screenshot required 5 minutes of prep: - Hide plugin admin menus - Clear admin notices - Remove admin bar items - Hide promotional banners - Clean up the interface **The Solution**: A Chrome extension that does it all with one click. **Technical Details**: - Built with Chrome Manifest V3 (a nightmare, but worth it) - Uses MutationObserver for dynamically loaded content - Saves preferences for consistent screenshots - Works with any WordPress site **Time Saved**: ~2 hours/week **GitHub**: [verygoodplugins/wp-clean-admin](https://github.com/verygoodplugins/wp-clean-admin) **Current Stars**: 7 ⭐ (we'll take it!)  *One click transforms cluttered WordPress admin into clean, screenshot-ready interface* ### Tool #2: FreeScout GPT Assistant - The Reply Robot **The Problem**: Writing "Thanks for reporting this! I've been able to reproduce the issue..." for the 10,000th time. **The Solution**: Chrome extension that integrates GPT-4 directly into FreeScout. **How It Works**: 1. Reads the entire conversation thread 2. Pulls customer data and history 3. Checks documentation for relevant info 4. Generates contextual responses in your tone 5. Inserts directly into the editor **The Magic**: It learns your style. My replies keep my "trademark overuse of emojis 🔥", while our other support tech stays properly professional. **Technical Challenges**: - FreeScout uses Summernote editor (nightmare to programmatically insert text) - Had to reverse-engineer their editor API - GPT-4 was too expensive, switched to GPT-4o for most tasks **Time Saved**: ~10 hours/week **Cost**: ~$4/month in OpenAI credits for 500 responses **GitHub**: [verygoodplugins/freescout-gpt-assistant](https://github.com/verygoodplugins/freescout-gpt-assistant) **Stars**: 11 ⭐  *AI-powered replies that understand context, customer history, and maintain your brand voice* ### Tool #3: FreeScout GitHub Integration - The Bug Tracker **The Problem**: Copy ticket to GitHub, lose formatting, forget to link back, manually update both systems forever. **The Solution**: FreeScout module with bidirectional GitHub sync. **Features**: - One-click issue creation with AI-generated descriptions - Automatic webhook syncing - Status updates flow both ways - PR links appear in tickets automatically **Smart Implementation**: - Uses GPT-5 models with new Responses API - Generation time cut from 20+ seconds to 10-15 seconds - Actually generates better descriptions than humans **What Almost Killed Us**: - SSL certificates on local dev (just disabled for .local domains, YOLO) - GitHub webhooks can randomly take 30 minutes to deliver - Teaching AI not to tag literally everything as "bug" **Time Saved**: ~5 hours/week **GitHub**: [verygoodplugins/freescout-github](https://github.com/verygoodplugins/freescout-github) **Stars**: 5 ⭐  *When everything works: One-click issue creation with perfect formatting and automatic linking*  *When webhooks hate you: The "why is nothing syncing?" debugging experience* ### Tool #4: MCP FreeScout Server - The Actual Magic 🪄 **The Problem**: Even with all the above tools, still too much clicking. **The Solution**: An MCP server that lets Claude or Cursor talk directly to FreeScout. **What This Enables**: Tell Cursor: "fix ticket #34811" and watch it: 1. Analyze the ticket 2. Create a worktree 3. Implement the fix 4. Create a PR 5. Draft a customer reply 6. Update ticket status All without touching anything. **Installation** (it's stupidly simple): ```bash npx @verygoodplugins/mcp-freescout@latest ``` **Time Saved**: ~8 hours/week **GitHub**: [verygoodplugins/mcp-freescout](https://github.com/verygoodplugins/mcp-freescout) **Stars**: 5 ⭐ > "The moment we knew we were living in the future: Watching Cursor fix a bug, create the PR, and write the customer reply without us touching anything. Still feels like cheating 🚀" --- ## 🤖 The Claude Automation Workflow ### The "Claude Implement" Magic This is where things get truly futuristic. We built a GitHub Actions workflow that lets Claude autonomously fix bugs. #### How It Works: 1. **Human adds label** - Support adds "claude implement" to a GitHub issue 2. **Claude wakes up** - Reads the issue, understands what's broken 3. **Creates a branch** - Following our naming conventions (feature/gh-issue-123-whatever) 4. **Implements the fix** - Actually writes working code 5. **Runs PHP CodeSniffer** - Enforces WordPress coding standards 6. **Creates draft PR** - With detailed description of what and why 7. **Comments on issue** - "Hey, I fixed this. PR is here. Please test." #### Success Rate: 70% Perfect on First Try The other 30%? Claude gets close enough that a human can finish in minutes instead of hours. #### Real Examples: - Fixed typos and documentation issues - Resolved complex WooCommerce integration bugs - Debugged a race condition that had been annoying us for months - Implemented new features based on customer requests ### Enter CodeRabbit - The Safety Net Letting AI write production code is terrifying. That's where CodeRabbit comes in ($12/month). **What CodeRabbit Does**: - Reviews every line of code - Catches security issues and bugs - Suggests improvements - Provides testing instructions - Generates understandable summaries **The Best Part**: Claude and CodeRabbit talk to each other! - Claude makes a PR - CodeRabbit reviews it - They discuss the best approach in comments - It's like watching two developers pair programming, except they're robots and never need coffee ### Support Agents Become QA Engineers Our support team can now: 1. See a bug in FreeScout → Click to create GitHub issue 2. Add "claude implement" label → Claude fixes it automatically 3. Read CodeRabbit's testing instructions → Clear steps on what to test 4. Use GitHub Updater plugin → Switch to PR branch with one click 5. Test the fix → Following CodeRabbit's guide 6. Report back → "✅ Works" or "❌ Still broken, here's why" **No Git knowledge required. No terminal. Just a dropdown menu.** --- ## 🚀 Setting Up Your Own AI Pipeline ### Phase 1: Foundation (Week 1) #### Step 1: Install Claude Desktop 1. Download from [claude.ai](https://claude.ai) 2. Get the Pro subscription (necessary for MCP servers) 3. Enable developer mode in settings #### Step 2: Configure Your First MCP Server Start with something simple like the filesystem server: ```json { "mcpServers": { "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/directory"] } } } ``` Save this as `~/Library/Application Support/Claude/claude_desktop_config.json`  *Claude Desktop settings showing MCP servers configured and ready for automation* #### Step 3: Test Basic Automation Ask Claude: "Create a daily standup template in my documents folder" If it works, you're ready for more complex integrations! ### Phase 2: Support Integration (Week 2) #### Step 1: Set Up FreeScout If you're not using FreeScout, adapt these concepts to your support tool. 1. Install FreeScout (self-hosted or cloud) 2. Get your API credentials 3. Install our MCP server: ```bash npx @verygoodplugins/mcp-freescout@latest ``` #### Step 2: Configure the Integration Add to your Claude Desktop config: ```json { "mcpServers": { "freescout": { "command": "npx", "args": ["-y", "@verygoodplugins/mcp-freescout"], "env": { "FREESCOUT_API_URL": "https://your-instance.freescout.net/api", "FREESCOUT_API_KEY": "your-api-key" } } } } ``` #### Step 3: Install Browser Extensions 1. [WP Clean Admin](https://github.com/verygoodplugins/wp-clean-admin) for screenshots 2. [FreeScout GPT Assistant](https://github.com/verygoodplugins/freescout-gpt-assistant) for AI replies Configure with your OpenAI API key. ### Phase 3: GitHub Automation (Week 3) #### Step 1: Install GitHub MCP Server ```bash npx @modelcontextprotocol/server-github@latest ``` #### Step 2: Set Up the Claude Implement Workflow 1. Copy our [workflow file](https://github.com/verygoodplugins/wp-fusion/blob/master/.github/workflows/claude-auto-implement.yml) 2. Add to `.github/workflows/` in your repo 3. Configure your GitHub secrets: - `ANTHROPIC_API_KEY` - `GITHUB_TOKEN` (with write permissions) #### Step 3: Install CodeRabbit 1. Go to [coderabbit.ai](https://www.coderabbit.ai) 2. Install the GitHub app 3. Configure for your repositories ### Phase 4: Advanced Integrations (Week 4) #### Install Additional MCP Servers Based on your needs: ```bash # Slack integration npx @modelcontextprotocol/server-slack@latest # PostgreSQL for database queries npx @modelcontextprotocol/server-postgres@latest # Browser automation npx @modelcontextprotocol/server-playwright@latest ``` #### Create Custom Workflows #### The Claude Automation Hub We've open-sourced our complete automation framework in the **[Claude Automation Hub](https://github.com/verygoodplugins/claude-automation-hub)** - a comprehensive toolkit that connects Claude Desktop to virtually any service using MCP (Model Context Protocol) servers.  **What's Included:** - **Slack Integration** - Full AI bot capabilities with workflow steps - **Browser Profile Management** - Persistent login sessions for web automation - **Node.js Version Management** - Automatic compatibility with NVM and other version managers - **Cloudflare Tunnels** - Secure, free tunneling for local development - **Pre-built Workflows** - Ready-to-use automation templates - **MCP Server Collection** - Integrations for WordPress, FreeScout, OpenMemory, and more **Sample Workflows:** **Morning Routine** (`workflows/daily/morning-routine.md`): - Calendar review and conflict detection - Email triage with AI categorization - Task prioritization based on deadlines - Focus block scheduling with distraction blocking **Weekly Review** (`workflows/weekly/review-planning.md`): - Automated metrics analysis across all platforms - Goal tracking with progress visualization - AI-assisted planning session with priority recommendations - Stakeholder report generation and distribution **Customer Success Check-in** (`workflows/on-demand/customer-success.md`): - Multi-channel customer history analysis (FreeScout, WhatsApp, Gmail) - Sentiment analysis of recent interactions - Personalized outreach message generation - Follow-up reminder scheduling --- ## 📊 Real Results & ROI ### The Numbers Don't Lie #### Before (July 2024): - **Average ticket resolution**: 4.2 hours - **Tickets needing GitHub**: ~40% - **Weekly admin time**: ~15 hours - **Customer satisfaction**: 4.2/5 - **Team morale**: 😞 #### After (November 2024): - **Average ticket resolution**: 1.1 hours (-74%) - **Tickets needing GitHub**: Still ~40% (but automated) - **Weekly admin time**: ~2 hours (-87%) - **Customer satisfaction**: 4.7/5 - **Bugs fixed by Claude**: ~70% success rate - **Team morale**: 🚀 ### Financial Breakdown #### Investment: - **Dev time**: ~160 hours ($8,000 at internal rate) - **Monthly costs**: - OpenAI API: ~$50 - Claude API: ~$30 - CodeRabbit: $36 (3 users) - GitHub Copilot: $19 - **Total**: $135/month #### Savings: - **Support time**: 60+ hours/month (~$3,000) - **Dev escalations**: ~20 hours/month (~$1,000) - **Total**: ~$4,000/month **Break-even**: 2 months **Annual net benefit**: ~$46,000 --- ## 💡 Lessons Learned & Best Practices ### What Went Hilariously Wrong #### Claude's Greatest Hits: - **The Overeager Refactor** - Asked Claude to fix a typo. It refactored the entire class "for better readability" 😅 - **The Infinite Loop Incident** - Claude and CodeRabbit got into an argument about code style. 47 comments later... - **The Helpful Deletion** - Claude decided some "unused" code was cluttering things up. It was not unused. - **Breaking Production on a Friday** - Support labeled an issue at 4:58 PM. Claude immediately "fixed" it. You know how this ends. #### Actual Disasters: - Chrome extension auto-update that hid ALL menus (Saturday morning emergency) - GitHub rate limits everywhere (cache everything or die) - MCP paths are lies (`${workspaceFolder}` doesn't mean what you think) - FreeScout updates breaking everything (now we lock versions) - Amazon SES approval took 5 attempts (FIVE!) ### Critical Success Factors #### 1. Start Small Don't try to automate everything at once. Pick one painful workflow and perfect it. #### 2. Always Create Draft PRs Claude is confident but not always right. Human review is essential. #### 3. Rate Limit Everything Claude will happily make 1000 API calls if you let it. #### 4. Add Kill Switches When robots go rogue, you need the big red button. #### 5. Never Deploy on Fridays Even if Claude says it's fine. Especially if Claude says it's fine. #### 6. Use TypeScript from Day One We're still porting things. Learn from our mistake. ### The Mental Model That Works #### Morning: Calendar First 1. Start with calendar review for context 2. Triage communications by priority 3. Set up environment for deep work 4. Create time blocks for focused tasks #### Processing: Aggregate Then Analyze 1. Gather data from multiple sources 2. Identify patterns and trends 3. Generate actionable insights 4. Store learnings for future reference #### Problem-Solving: Research Before Implementing 1. Analyze issue thoroughly first 2. Check documentation and past solutions 3. Create isolated development environment 4. Document solution for knowledge base --- ## 📚 Resources & Next Steps ### Essential Links #### Our Tools (All Free & Open Source): - **[Claude Automation Hub](https://github.com/verygoodplugins/claude-automation-hub)** - Complete MCP-based automation framework with Slack integration, browser profiles, and workflow templates - [WP Clean Admin](https://github.com/verygoodplugins/wp-clean-admin) - Screenshot tool - [FreeScout GPT Assistant](https://github.com/verygoodplugins/freescout-gpt-assistant) - AI replies - [FreeScout GitHub Module](https://github.com/verygoodplugins/freescout-github) - Two-way sync - [MCP FreeScout Server](https://github.com/verygoodplugins/mcp-freescout) - Complete automation #### Must-Watch & Must-Read: - **[WP Product Talk Episode - AI Workflows](https://www.youtube.com/watch?v=CVHlPDXGMmY&t=3119s)** - Complete walkthrough with live demos - [We Built 4 Tools to Fix Our Support Pipeline](https://wpfusion.com/business/we-built-4-tools-to-fix-our-support-pipeline-and-you-can-have-them-for-free/) - [How to Run Claude Code on a Server](https://jewelhuq.medium.com/how-to-run-claude-code-on-a-server-and-access-it-anywhere-c31b31488ed9) - [6 MCP Servers Every Developer Needs](https://medium.com/coding-nexus/6-mcp-servers-every-developer-needs-to-try-622b3e639403) #### MCP Resources: - [Official MCP Servers](https://github.com/modelcontextprotocol/servers) - [MCP Documentation](https://modelcontextprotocol.io) - [Community MCP Servers](https://github.com/topics/mcp-server) ### Your 30-Day Implementation Plan #### Week 1: Foundation - [ ] Install Claude Desktop - [ ] Configure first MCP server - [ ] Run first automation - [ ] Document time saved #### Week 2: Support Integration - [ ] Set up support tool integration - [ ] Install browser extensions - [ ] Create first AI reply template - [ ] Train team on new tools #### Week 3: Development Automation - [ ] Configure GitHub integration - [ ] Set up Claude implement workflow - [ ] Install CodeRabbit - [ ] Test first automated fix #### Week 4: Optimization - [ ] Add custom workflows - [ ] Fine-tune prompts - [ ] Set up monitoring - [ ] Calculate ROI ### Join the Community #### WP Product Talk: - Website: [wpproducttalk.com](https://wpproducttalk.com) - Twitter: [@WPProductTalk](https://twitter.com/wpproducttalk) - Watch episodes: [YouTube](https://youtube.com/@wpproducttalk) #### Very Good Plugins: - Website: [verygoodplugins.com](https://verygoodplugins.com) - GitHub: [github.com/verygoodplugins](https://github.com/verygoodplugins) - Blog: [wpfusion.com/blog](https://wpfusion.com/blog) ### Get Help - **Questions?** Open an issue on any of our repos - **Want to contribute?** PRs welcome! - **Need consulting?** Contact us through our website --- ## 🎯 The Bottom Line We spent August building tools to automate our support pipeline. Then we spent September teaching robots to fix bugs. Now our support team spends most of their time actually helping customers instead of doing admin work. **Cost**: ~$8,000 in dev time **Savings**: ~$4,000/month **Real Win**: Support doesn't suck anymore Bugs get fixed while we're eating lunch. Our team is happier. Customers get fixes faster. Is it perfect? Hell no. Claude occasionally does something catastrophically stupid. CodeRabbit sometimes misses obvious bugs. The whole system falls apart if GitHub is down. But you know what? **It's still 100x better than doing everything manually.** We're living proof that a small team with some questionable ideas and too much caffeine can build something that actually works. Sort of. Most of the time. ### Remember: > "The best automation is the one you actually use. Start small, iterate often, and compound your time savings!" --- ## ⚠️ Disclaimer If you implement this and Claude accidentally deletes your production database, that's on you. But definitely tell us about it, we need a good laugh. Seriously though, if you build something cool with these tools, let us know. Nothing makes us happier than seeing our weird experiments actually help people. Even if those people are robots. --- *P.S. - This entire README was written with Claude's help. It took 5 minutes instead of 2 hours. The irony is not lost on us. 🤖*