n8n-mcp.
An open-source tool that connects AI agents to the n8n automation platform and lets them build working workflows — 21,600 GitHub stars, 100,000 users, 19,000 SaaS users. Used by engineering teams at Deutsche Telekom, PayPal, Mercado Libre, MIT and NYU, among others.
- Industry
- Open source / own product
- Timeline
- 2024–present
- Outcome
- 21,600 ★ · 100,000 users
The problem
n8n is one of the most popular process-automation platforms. AI agents — like Claude — could in theory build workflows in it, but in practice they were guessing: they didn't know the current documentation, they got configurations wrong, and nothing checked whether what they built would actually run. The layer that gives agents reliable knowledge of the platform and verifies their work was missing.
What I built
n8n-mcp connects AI agents to n8n — but it's not just a "cable". It's a full service layer that makes agents genuinely capable of building working automations:
- knowledge instead of guessing — the agent has access to the full, up-to-date documentation of n8n nodes and checks the facts before it builds anything,
- output validation — configurations and entire workflows are verified automatically before they reach production,
- workflow operations — the agent creates, edits and tests workflows directly on the n8n instance.
On top of that: a self-fine-tuned model for workflow generation, a cache covering several hundred thousand patterns, and a commercial SaaS platform. The whole thing — code, infrastructure and platform — was built agentically (Claude Code + MCP).
The outcome
Used by engineering teams at Deutsche Telekom, T-Systems, PayPal, Mercado Libre, Intercom, Replit, MIT and NYU, among others. A participant in the GitHub Secure Open Source Fund (edition 4, 2026). Working proof that agentic software development scales to production infrastructure.