n8n-mcp.
An open-source tool that connects AI agents to the n8n automation platform and lets them build automations that actually run — 22,000+ GitHub stars, 100,000+ users, 1.1 million+ workflows built. Used by engineering teams at Deutsche Telekom, PayPal, Mercado Libre, MIT and NYU, among others.
- Industry
- Open source / own product
- Timeline
- 2024–present
- Outcome
- 100,000+ users · 1.1M+ workflows built
The problem
Automation platforms like n8n can take an enormous amount of repetitive work off a team's plate — but building and maintaining the automations takes expertise that's scarce and expensive. AI agents promised to close that gap and build the automations themselves. In practice they guessed: they didn't know the platform's current capabilities, they got the configuration wrong, and nothing checked whether what they produced would actually run. "Let AI build your automation" wasn't something you could trust in production.
What I built
n8n-mcp is the layer that makes AI agents genuinely capable of building working automations — not a thin connector, but the knowledge and the guardrails an agent needs:
- Knowledge instead of guessing — the agent works from complete, always-current knowledge of the entire n8n platform — every one of its 1,800+ building blocks — and checks the facts before it builds anything.
- It checks the work before it ships — individual settings and whole automations are verified automatically, so mistakes are caught before they reach production, not after.
- It acts, not just advises — the agent builds, edits and tests automations directly on the live system.
- It works inside the tools people already use — Claude, Cursor, VS Code and others, so there's nothing new to learn.
- Free and open, or hosted with zero setup — anyone can run it for free, or use the hosted version (with a free tier) and skip the infrastructure entirely.
Behind it sit a model I fine-tuned myself for generating workflows and a library of hundreds of thousands of real-world patterns. The whole thing — the tool, the infrastructure and the commercial platform — was built agentically, with Claude Code.
The outcome
n8n-mcp is used by more than 100,000 people, including engineering teams at Deutsche Telekom, T-Systems, PayPal, Mercado Libre, Intercom, Replit, MIT and NYU. It has handled close to 25 million agent requests with a 99.5% success rate, and more than 1.1 million workflows have been built with it. It's also the engine behind the self-service automation teams I set up for clients — the reason someone without specialist skills can describe what they need and get a working automation back. Proof, at the scale of a real product, that agentic software development holds up in production.