ADK Workflow Wizard

What is this project?

A sophisticated multi-agent system built with Google's Agent Development Kit (ADK) that automatically generates production-ready n8n workflows from natural language requirements. The system uses specialized AI agents working in parallel to plan, develop, and document n8n workflows with real-time Context7 verification.

Why did I build it?

I wanted to explore Google's Agent Development Kit and see how multi-agent systems could automate complex workflow generation. Creating n8n workflows requires understanding node types, connections, and best practices—perfect for testing how agents can collaborate using real-time documentation verification through Context7 MCP.

How does it work?

The system uses a multi-agent architecture:

1. Workflow Router: Routes user requirements to appropriate agents
2. Workflow Planner: Acts as technical architect, using Context7 MCP to verify n8n node types and design optimal flows through iterative conversation
3. Developer Agent Team: Three specialized agents work in parallel:
   • Trigger Generator: Creates trigger and data collection nodes
   • Processing Generator: Builds transformation and logic nodes
   • Output Generator: Handles notifications and error handling
4. Workflow Compiler: Merges all sections into the final workflow JSON
5. Documentation Agent: Generates user-friendly setup guides automatically

All agents have access to Context7 MCP for real-time n8n documentation verification, ensuring generated workflows use valid node types and configurations.

Architecture Diagram

Tech Stack

• Google Agent Development Kit (ADK): Core multi-agent framework
• Context7 MCP: Real-time n8n documentation verification
• Opik: Observability, tracing, evaluation, and prompt optimization
• Gemini 2.5 Flash: LLM model via OpenRouter
• Python: Implementation language

What were the biggest challenges?

• Designing agent boundaries and collaboration patterns to avoid loops while ensuring thorough planning
• Integrating Context7 MCP across multiple agents for consistent documentation verification
• Coordinating parallel agent execution (triggers, processing, output) into a coherent workflow
• Ensuring generated n8n JSON is valid and immediately importable
• Balancing detailed planning with efficient execution time

How can you use it or build on this?

• Generate n8n workflows from natural language descriptions
• Learn multi-agent system design patterns with Google ADK
• Extend with more specialized agents for specific n8n node types
• Use as a template for other workflow automation tools
• Explore Context7 MCP integration for real-time documentation access

Links

• GitHub: https://github.com/Folken2/adk-wizard

What's next?

• Add support for more complex n8n workflow patterns
• Integrate workflow testing and validation before output
• Expand agent capabilities for error handling and edge cases
• Add visual workflow preview generation
• Enhance documentation generation with troubleshooting guides

In the age of AI, stay curious!