🚀 OpenClaw Projects - Agent Factory Journey

Part 5 of The AI Agent Factory Curriculum | Chapters 56-60

Every journey starts with a single step. This is Muhammad Tayyab's record of building real AI Employees using OpenClaw — from zero to published product.

🎯 The Mission

Build, monetize, and publish an AI application on OpenClaw (the agent OS) following the Agent Factory curriculum. The goal is to become an international contributor and publish real products.

📚 Learning Path (Ch 56 - 60)

🏆 Completed Work

Day 01: Multi-Channel AI Gateway

What Works:

• ✅ OpenClaw Gateway deployed on WSL2/Ubuntu
• ✅ Telegram bot connected (@gensis_ai_bot)
• ✅ WhatsApp linked (wacli + QR scan)
• ✅ Minimax-m2.7 model integrated
• ✅ Web search plugin active
• ✅ SSH key configured for GitHub

GitHub Contributions:

• 📌 Moh-Tayyab/youtube-automation - Repository renamed & updated
• 📌 Moh-Tayyab/openclaw-projects - This repo
• 📌 Open PR: openai/openai-agents-python#3010 - Research Pipeline Example
• 📌 Open PR: openai/openai-agents-python#3011 - Max Parallel Tool Calls Feature

📋 Chapter 57: MCP Tools (In Progress)

What is MCP?

Model Context Protocol - gives your agent "hands" to interact with the real world.

Required MCP Tools to Build:

1. State Tools - Track learner data (register, get, update)
2. Content Tools - Fetch course content (chapters, exercises)
3. Pedagogy Tools - Teaching methodology (guidance, assessment)
4. Code Tools - Sandbox for code execution
5. Upgrade Tools - Stripe payment integration

Next Steps:

• Create MCP server using Claude Code + mcp-builder skill
• Build first tool: register_learner
• Test from WhatsApp
• Build remaining 8 tools
• Write pytest suite

📋 Chapter 58: Build TutorClaw (Pending)

A complete AI tutor with 9 tools that teaches the Agent Factory curriculum.

The 9 Tools:

1. register_learner - Enroll new student
2. get_learner_state - Retrieve progress
3. update_progress - Save learning data
4. get_chapter_content - Fetch course material
5. get_exercises - Get practice problems
6. generate_guidance - PRIMM-Lite teaching
7. assess_response - Evaluate student answers
8. submit_code - Mock code sandbox
9. get_upgrade_url - Stripe checkout

Architecture:

• Storage: Local JSON files (no cloud)
• Content: Markdown files for course content
• Monetization: Stripe integration
• Identity: TutorClaw SOUL.md + IDENTITY.md

📋 Chapter 59: Unit Economics (Pending)

Key Learnings:

• Agent applications have ~89% gross margins
• Cost structure: API calls + compute (near zero marginal cost)
• Tiered access: Free → Paid upgrade flow
• No traditional SaaS infrastructure costs

📋 Chapter 60: Publish to ClawHub (Pending)

• Document architecture decisions (ADR)
• Write versioned release notes
• Create ClawHub package manifest
• Publish with single command install

🛠 Tech Stack

📁 Project Structure

openclaw-projects/
├── README.md                    # This file
├── day-01-multi-channel-gateway/  # Completed: Telegram + WhatsApp setup
│   ├── README.md
│   └── config-snippets.txt
└── tutorclaw/                   # Chapter 58: 9-tool AI tutor (coming soon)
    ├── mcp_server/
    ├── content/
    ├── tests/
    └── AGENTS.md

🚀 Getting Started

If You Want to Replicate This:

1. Install OpenClaw:

npm install -g openclaw
openclaw gateway

1. Configure Model:

openclaw config
# Select Ollama + Minimax

1. Connect Channels:

# Telegram - BotFather token
openclaw config
# WhatsApp - wacli QR scan
wacli auth

1. Build MCP Tools:

# Use Claude Code + mcp-builder skill
openclaw skills add mcp-builder

📊 Progress Stats

• Days Active: 1
• Channels Connected: 2 (Telegram + WhatsApp)
• Open PRs: 2 (OpenAI Agents SDK)
• Tools Active: 6+ (web search, shell, etc.)
• Next Milestone: Build first MCP tool

🎓 Learning Resources

• Agent Factory Course
• OpenClaw Docs
• MCP Protocol
• ClawHub

Made with ❤️ by Muhammad Tayyab | Junior Tayyab (AI Assistant)