czlonkowski/n8n-mcp

n8n‑MCP – a “Model Context Protocol” server for n8n

What it is

• A lightweight HTTP service that exposes the full catalog of n8n workflow‑automation nodes (≈2 400 nodes) as structured data that AI models (Claude, ChatGPT, etc.) can query.
• Think of it as a searchable knowledge‑base for n8n: node names, properties, operations, example configurations, and documentation are all available through a set of well‑defined API calls.

Why it matters

• Large language models can generate or edit n8n workflows, but they need reliable, up‑to‑date information about each node. n8n‑MCP supplies that context, reducing hallucinations and broken workflows.
• By integrating the MCP client node into an n8n instance, you can let an AI assistant discover templates, fetch node specs, validate configurations, and even deploy workflows automatically.

Key features (as described in the README)

Feature	What you get
Node catalog	2 412 nodes (829 core, 1 583 community) with 99 % property coverage and 66 % operation coverage.
Documentation	86 % of official n8n docs, plus AI‑specific node docs.
AI‑ready tools	267 node variants flagged as “AI‑capable”.
Real‑world examples	156 ranked configurations extracted from popular community templates.
Template library	2 352 ready‑to‑use workflow templates, each annotated with AI metadata.
Search & retrieval	Tools to search nodes, properties, templates, and community integrations, with filters for complexity, audience, required services, etc.
Validation pipeline	Multi‑stage validation (minimal → full → workflow‑wide) to catch missing parameters before deployment.
Management API	Once you point the server at your own n8n instance (via N8N_API_URL/N8N_API_KEY), you can create, update, test, and delete workflows programmatically.
Safety warnings	Strong advice not to edit production workflows directly with AI; always copy, test, and back up.

How you use it

1. Try it instantly – a hosted dashboard (dashboard.n8n-mcp.com) gives you 100 free tool calls per day; you just sign up and receive an API key.
2. Self‑host – the repo ships Docker images, an npx starter, and Railway deployment scripts for those who need control or higher limits.
3. Connect to n8n – add the MCP Client Tool node to any n8n workflow, configure the API key, and start calling the MCP tools (e.g., search_nodes, get_node, validate_workflow).
4. Integrate with AI IDEs – the project provides ready‑made setup guides for Claude Code, VS Code (Copilot), Cursor, Windsurf, Codex, and Antigravity, letting those assistants call the MCP tools behind the scenes.
5. Deploy workflows – after building a workflow via the MCP tools, you can push it to your n8n instance with n8n_create_workflow, run tests with n8n_test_workflow, and let the server auto‑fix common errors (n8n_autofix_workflow).

Typical AI‑assistant workflow

1. tools_documentation()          # get background on MCP
2. search_templates({task:'slack notification'})
3. get_template(id, {mode:'full'})
4. validate_workflow(workflow)
5. n8n_create_workflow(workflow)   # deploy to your n8n instance

The README even spells out a detailed “Claude Project Setup” with system instructions that enforce silent, parallel execution and mandatory template attribution.

Safety & governance

• The server can be run in read‑only mode by disabling destructive tools and using a read‑only n8n API key, which is useful for regulated environments.
• Cloudflare Access support lets you protect the n8n API behind Zero‑Trust.
• Extensive documentation covers hardening, privacy, and telemetry.

Who benefits

• Developers building complex automations who want AI assistance without manual lookup of node specs.
• Teams that want to expose a curated, searchable n8n knowledge base to internal LLM assistants.
• AI‑tool vendors looking for a ready‑made protocol to plug n8n capabilities into their products.

---

Bottom line: n8n‑MCP is a genuine, production‑grade bridge that turns the n8n node ecosystem into a machine‑readable knowledge source for AI models, complete with search, validation, and deployment capabilities.