Skip to main content
Back

n8n-mcp Hits 22K Stars Fixing AI Agents That Hallucinate Broken Workflows

A community-built MCP server hands Claude, Cursor and Windsurf deep knowledge of 2,541 n8n nodes so agents can design, validate and deploy real workflows.

Repo, Full Access
  • n8n-mcp is an MIT-licensed MCP server exposing 2,541 n8n nodes to Claude, Cursor and Windsurf.
  • Provides seven core tools for node search, docs, validation and template discovery across 2,352 workflows.
  • Sixteen management tools cover create, update, autofix, execute and audit against a live n8n API.
  • Multi-level validation ladder catches misconfigurations before deployment, cutting runtime failures from default parameters.
  • Hosted tier at dashboard.n8n-mcp.com gives 100 free calls/day; self-host via npx or Docker for unlimited usage.
  • Companion n8n-skills plugin adds seven Claude Code skills for expressions, patterns and error recovery.

Building n8n workflows by hand involves a lot of clicking. Building them with a generic LLM involves a lot of hallucinated node names, wrong parameters, and broken JSON exports. A community project called n8n-mcp has crossed 22,000 stars by fixing the second problem: it gives coding agents a structured, always-current view of every n8n node so they can build workflows that actually run.

n8n-mcp is a Model Context Protocol (MCP) server that provides AI assistants with comprehensive access to n8n node documentation, properties, and operations. Under the hood it ships a SQLite database describing n8n's 2,541 workflow automation nodes (832 core plus 1,709 community), along with a library of validated workflow templates the model can copy from instead of guessing.

Why generic coding agents keep failing at n8n

n8n has hundreds of integrations, each with dozens of properties and operation-dependent required fields. When you ask Claude or Cursor to "build a webhook to Slack workflow," the model tends to invent parameter names, forget required fields, or emit workflow JSON that n8n rejects on import. The n8n-mcp maintainer describes the failure modes bluntly: using MCP tools incorrectly, getting stuck in validation error loops, not knowing which workflow patterns to use, and misconfiguring nodes and their dependencies.

The server tackles this by exposing a small, opinionated tool surface that forces the agent through a discovery, validation, and deployment loop instead of letting it free-associate JSON.

What the tool surface actually looks like