N8n MCP

What AutomateLab-tech/n8n-mcp MCP server does

AutomateLab-tech/n8n-mcp MCP server gives MCP clients focused tools for building and troubleshooting n8n workflows. Its stateless operations accept descriptions or JSON directly, so they can generate a workflow, inspect its structure, compare two versions, or analyze an execution without connecting to n8n. The server also provides live-instance operations for listing, reading, creating, activating, and inspecting workflows and executions.

Workflow generation supports webhook, schedule, and RSS triggers; common services such as Slack, Google Sheets, Discord, Gmail, Notion, and HTTP; and AI Agent arrangements with LangChain agent, chat model, memory, and optional HTTP tool connections. Generated workflows include node IDs, positions, connections, and node type versions.

How it works

The tools are exposed over MCP using dot-notation groups for nodes, workflows, and executions. node_scaffold produces one TypeScript file containing an INodeType skeleton for a custom n8n node. It does not write files or make network requests. workflow_lint applies deterministic checks for issues such as deprecated node types, missing credentials, invalid connections, absent type versions, expression references to missing nodes, and several n8n-specific configuration risks.

For debugging, provide execution JSON from the n8n interface or from the executions API. execution_explain summarizes each node, identifies zero-item outputs and unresolved expressions, and reports errors with hints. execution_timeline instead presents timing, item counts, and error status in a markdown table. execution_replay creates a smaller workflow containing a Manual Trigger, pinned seed data, and the selected target node.

workflow_diff compares workflows semantically, including node fields, connections, and settings, while ignoring small position changes and timestamp metadata. The live tools use the n8n API and can be combined with generation, linting, and explanation for describe-to-deploy or deployed-workflow audit flows.

Setup and configuration

Install the AutomateLab-tech/n8n-mcp MCP server with Node.js 20 or later. A local MCP host can run npx -y @automatelab/n8n-mcp. The README provides configuration examples for Claude Desktop and Cursor, including the server command and optional environment variables.

The stateless tools do not need n8n credentials. Set N8N_API_URL and N8N_API_KEY to enable live-instance tools. The API key is created in n8n under Settings, API, and Create API key. Runtime policy variables can restrict access: N8N_MCP_READ_ONLY disables selected mutating tools, N8N_MCP_DISABLED_TOOLS prevents named tools from registering, and workflow ID or tag allowlists constrain REST operations.

Tools and capabilities

• Generate workflow JSON from plain-English requirements.
• Scaffold a TypeScript custom-node implementation.
• Lint workflow structure and n8n compatibility issues.
• Compare workflow versions and report field-level changes.
• Explain failed executions and render execution timelines.
• Build single-node replay workflows with pinned input items.
• List, fetch, create, activate, and deactivate live workflows.
• List recent executions with optional full execution data.

Limitations and notes

Generated workflow output is non-deterministic because node IDs and webhook paths are randomized. The server generates code and JSON but does not write files for node scaffolding. Live operations require access to an n8n API and the two n8n credentials described above; stateless analysis cannot retrieve data from a running instance by itself.

The AutomateLab-tech/n8n-mcp MCP server uses dot-notation tool names in the current release. Older prompts or scripts using the former n8n_* names need updating. Creating a workflow leaves it inactive, so activation is a separate operation. The server is released under the MIT license.