P58-n8n — Path58's n8n MCP Server

Enterprise-grade n8n workflow validation for LLMs

P58-n8n gives your AI assistant deep knowledge of n8n — 1,982 nodes, 40,179 operations, and 26 auto-fix rules — so it can build, validate, and fix workflows correctly the first time.

Why P58-n8n?

| Metric | P58-n8n | n8n-mcp | Claude alone | |--------|---------|---------|-------------| | Accuracy | 63.6% | 59.1% | 42.0% | | Cost per prompt | $0.25 | $0.76 | ~$0.05 | | Reliability | 100% | 82% | 100% | | Token usage | 1x | 3.2x | 0.3x |

4.5 percentage points more accurate, 67% cheaper, and 100% reliable. Benchmarked on 22 real-world workflow generation tasks.

Quick Start (< 5 minutes)

1. Configure your AI client

Claude Code (one command):

claude mcp add p58-n8n -- npx -y @path58/n8n-mcp

Claude Desktop:

Add to your claude_desktop_config.json (find your config path):

{
  "mcpServers": {
    "p58-n8n": {
      "command": "npx",
      "args": ["-y", "@path58/n8n-mcp"]
    }
  }
}

Then restart Claude Desktop.

Gemini CLI, global install, and other options: docs/INSTALLATION.md

2. Try it

Ask your AI assistant:

"What nodes does n8n have for sending Slack messages?"

or

"Validate this n8n workflow JSON for me: { ... }"

8 MCP Tools

Validation (4 tools)

| Tool | What it does | |------|-------------| | validate_workflow | Run L1-L6 validation on workflow JSON — structure, nodes, credentials, connections, parameters, and patterns | | get_operation_schema | Get exact parameter requirements for any node operation — prevents hallucinated parameters | | check_parameter | Validate a specific parameter value in real-time — type checking, enum validation, typo detection | | suggest_fix | Get concrete fix suggestions for validation issues — rename, add, remove, or replace |

Discovery (4 tools)

| Tool | What it does | |------|-------------| | list_nodes | Browse and search 1,982 n8n nodes by category or keyword | | list_operations | List all operations for a node type, grouped by resource | | get_node_info | Get comprehensive node details — operations, credentials, properties | | find_similar_pattern | Search 9,000+ validated workflows for working examples |

Full reference with input/output schemas: docs/TOOLS.md

What makes P58-n8n different

Ground-truth catalog. Every parameter, operation, and connection rule comes from the actual n8n node catalog — not LLM guesses.

6-level validation. L1 (structure) through L6 (flow patterns) catches problems before execution:

L1 Structure  →  L2 Nodes  →  L3 Credentials  →  L4 Connections  →  L5 Parameters  →  L6 Patterns
   ~30ms          ~60ms         ~80ms              ~180ms             ~850ms            ~200ms

Full pipeline: ~1.4 seconds.

26 auto-fix rules. Automatic repair for common issues — orphan connections, deprecated nodes, missing parameters, dead-end flows, and more.

9,000+ validated workflows. Real working examples your AI can reference when building new workflows.

Usage Examples

See docs/EXAMPLES.md for detailed walkthroughs:

1. Discover & Build — Find the right nodes and build a Slack notification workflow
2. Validate & Fix — Catch and repair issues in an existing workflow
3. Schema-Driven Generation — Use operation schemas to generate correct parameters

Architecture

P58-n8n runs as a local stdio MCP server. No API keys or cloud services required.

AI Client (Claude/Gemini)
    ↕ MCP Protocol (stdio)
P58-n8n Server
    ↕
Validation Engine (L1-L6) + Node Catalog (1,982 nodes, 40,179 ops)

See docs/ARCHITECTURE.md for the full system design.

Troubleshooting

Server not starting

# Verify Node.js 18+
node --version

# Test the server directly
npx -y @path58/n8n-mcp
# Should start without errors (it waits for MCP input on stdio)

Tools not appearing in Claude Desktop

1. Restart Claude Desktop after editing config
2. Check config file path is correct for your OS
3. Verify JSON syntax — trailing commas break it

"Command not found" errors

# Ensure Node.js and npx are available
which npx

# Or install globally instead
npm install -g @path58/n8n-mcp
p58-n8n  # Run directly

ChatGPT Desktop

ChatGPT Desktop is not supported — it requires HTTP/SSE transport, but P58-n8n uses stdio. See docs/INSTALLATION.md for details.

Contributing

P58-n8n is currently in soft launch (friends & family). Feedback welcome:

• Issues: GitHub Issues
• Email: [email protected]

License

MIT — see LICENSE for details.

Built by Path58