Earlier, MCP could only run workflows. Now it can actually:
- Create workflows
- Update workflows
This changes how we build automations.
💡 Quick context:Before this, many of us were using the community MCP (czlonkowski/n8n-mcp) to build workflows with AI.
Here’s what’s better now:
- The AI no longer gets overloaded with all node dataIt only fetches what’s needed → better context, better output
- No more guessing parameters or broken JSONIt now uses exact definitions → more reliable workflows
- Setup is much simplerNo Docker, no extra service → just enable it in n8n
⚠️ Still some limitations:
- You need to manually add credentials
- Workflows are ~80% ready (you refine the rest)
- Needs public access for tools like Claude / Cursor
Overall, this is a big step toward building n8n workflows directly from AI tools without jumping back and forth.
Curious — has anyone here tried the new MCP yet?Would love to hear your experience 👇