| name | n8n-mcp |
| description | Use when building, reviewing, debugging, validating, or safely deploying n8n workflows with the czlonkowski/n8n-mcp MCP server. Trigger on requests like "n8n workflow 만들어줘", "n8n-mcp 연결", "n8n 노드 찾아줘", "n8n 자동화 설계", "n8n credential/workflow 관리", or any task where an agent needs n8n node docs, templates, workflow JSON validation, or n8n API management. Always prefer read-only documentation/search first, work on copied workflows before production edits, and validate before deploy. |
n8n MCP
czlonkowski/n8n-mcp turns n8n's node catalog, templates, workflow validation, and optional n8n instance API into MCP tools. Treat it as two things:
- Library brain: search node docs, properties, operations, examples, and templates without touching an n8n instance.
- Workflow hands: if
N8N_API_URL and N8N_API_KEY are configured, create/update/test workflows against a real n8n instance.
The safety model is simple: docs-first, copy-first, validate-before-deploy. Never let an AI edit a production workflow directly as the first move.
When this runs
Use this skill when the user asks for any of these:
- Connecting
n8n-mcp to an agent or MCP client.
- Designing or generating n8n workflows.
- Finding the right n8n node, operation, property, or credential schema.
- Validating, autofixing, importing, deploying, or testing workflow JSON.
- Managing n8n workflows through API-backed MCP tools.
- Building automation around webhooks, schedules, AI agent nodes, Slack/Telegram/Discord, Google Workspace, Notion, Airtable, databases, or HTTP APIs in n8n.
Do not use this as a generic n8n tutorial if no MCP/tooling is involved. Do not use it to directly mutate production workflows without backups and confirmation.
Quick setup for Hermes Agent
For docs/template-only use, no n8n API credentials are required:
hermes mcp add n8n-mcp \
--command npx \
--env MCP_MODE=stdio LOG_LEVEL=error DISABLE_CONSOLE_OUTPUT=true \
--args n8n-mcp
Verify and reload:
hermes mcp list
hermes mcp test n8n-mcp
Inside a running Hermes session or gateway:
/reload-mcp
If tools still do not appear, start a new session or restart the gateway:
/restart
Setup with n8n API management
Aaron's default n8n instance is:
https://n8n.restack.tech
Only add API access when the user explicitly wants the agent to manage an n8n instance:
hermes mcp add n8n-mcp \
--command npx \
--env MCP_MODE=stdio LOG_LEVEL=error DISABLE_CONSOLE_OUTPUT=true N8N_API_URL=https://n8n.restack.tech N8N_API_KEY=*** \
--args n8n-mcp
Avoid putting real API keys directly into shell history when possible. Prefer whatever secret mechanism the current environment supports, then pass the values to the MCP server from that secret source.
For local n8n instances:
N8N_API_URL=http://localhost:5678
For Docker-to-host access, n8n-mcp docs often use:
N8N_API_URL=http://host.docker.internal:5678
If the server blocks localhost/private-network calls through an SSRF guard, check the upstream self-hosting docs for WEBHOOK_SECURITY_MODE=moderate or equivalent localhost allowance before debugging the workflow itself.
Operating procedure
1. Classify the task by risk
- Read-only: node search, property docs, template discovery, JSON explanation. Safe to proceed.
- Drafting: generate workflow JSON or a plan. Safe if no production write happens.
- Staging write: create/update a copied workflow or dev instance. Proceed after explaining target and backup.
- Production write/delete/credential change: ask for explicit confirmation, make/export a backup first, then validate and test.
2. Start with docs and examples
Before assembling JSON from memory, use MCP docs/search tools to identify:
- exact node type names;
- operation/resource names;
- required properties;
- credential requirements;
- example configurations;
- known validation constraints.
n8n node names and schemas change over time. The MCP catalog is the source of truth for node details.
3. Build the workflow in small slices
For non-trivial workflows, design by vertical slice:
- Trigger node.
- One data source or input normalizer.
- One transformation/branch.
- One output/action.
- Error handling and observability.
After each slice, validate the partial workflow JSON. Do not generate a huge workflow and only validate at the end.
4. Validate before deploy
Use available n8n-mcp validation/autofix tools before any real activation:
- validate node-level configuration;
- validate full workflow connections;
- check required credentials are placeholders or bound correctly;
- run autofix only on copied/dev workflows unless the user approves production changes;
- test webhook/schedule/manual triggers in a safe environment.
5. Preserve production workflows
Before modifying an existing workflow:
- Fetch/export the current workflow.
- Save the backup artifact path or workflow version ID.
- Clone/copy it when possible.
- Apply changes to the copy/dev workflow first.
- Validate and test.
- Only then update production with explicit user approval.
Common tool patterns
Actual tool names may vary as n8n-mcp evolves. Look for these capabilities:
- Search nodes by task or keyword.
- Get node details, properties, operations, and examples.
- Validate a node's minimal or full configuration.
- Search templates by nodes, task, metadata, service, or audience.
- Fetch a template in
nodes_only, structure, or full mode.
- Create/list/get/update/delete n8n workflows when API credentials are configured.
- Validate or autofix an existing n8n workflow by ID.
- Test/trigger a workflow and inspect executions.
- List credential schemas or manage credentials when explicitly requested.
Workflow design checklist
When creating a workflow, capture these before writing JSON:
- Trigger: webhook, schedule, manual, chat, form, app event, or polling.
- Inputs: schema, example payload, auth, pagination, rate limits.
- Transformations: mapping, filtering, splitting, merging, deduplication.
- Outputs: destination system, idempotency key, retry behavior.
- Credentials: which services need credentials, and whether placeholders are acceptable.
- Failure behavior: error workflow, Slack/Telegram alert, dead-letter storage, retry/backoff.
- Observability: execution names, tags, log nodes, audit trail.
- Environment: dev/staging/prod target, activation state, webhook URL.
Safety guardrails
- Never edit production workflows directly with AI. Copy first.
- Never invent credential IDs. Ask for them or discover them through approved read-only credential listing.
- Never paste API keys or secrets into workflow JSON unless the user explicitly chooses that pattern. Prefer n8n credentials.
- Never activate a workflow until validation passes and the user approves the target.
- Never delete workflows or credentials without explicit confirmation and a backup.
- Treat webhook URLs, execution payloads, and credentials as sensitive.
- If the MCP server only has docs mode, do not claim that a workflow was deployed or tested against n8n.
Local troubleshooting
If MCP startup fails:
- Confirm Node/npm/npx exists:
node --version
npm --version
npx --version
- Test the package directly:
MCP_MODE=stdio LOG_LEVEL=error DISABLE_CONSOLE_OUTPUT=true npx n8n-mcp
- If JSON-RPC parsing errors appear, ensure
MCP_MODE=stdio and quiet logging env vars are set.
- If n8n API tools fail, verify
N8N_API_URL, N8N_API_KEY, network access, and n8n API permissions.
- If local n8n is blocked, check upstream docs for localhost/SSRF settings.
Verification checklist
