Node validation checks one node against what that node actually requires; workflow validation then checks the graph around it — connections, expressions, AI tool wiring — and returns fixes rather than a verdict. Template search means the starting point is usually somebody's working workflow instead of an empty canvas.
An n8n knowledge, validation and management service. Seven core tools cover the node reference, the template library and validation, and need no n8n instance at all. Sixteen more manage a live instance — creating, updating, deploying, testing and auditing workflows — and switch on once `N8N_API_URL` and `N8N_API_KEY` are configured.
- N8n nodes searched full-text, across verified and community sources, with real configuration examples on request (`search_nodes`)
- One node read at the depth you need — minimal, standard or full — plus a docs mode, a property search, and version modes for comparing, spotting breaking changes and reading migrations (`get_node`)
- A node configuration validated quickly for required fields, or fully against one of four profiles: minimal, runtime, ai-friendly or strict (`validate_node`)
- A whole workflow validated, including the AI Agent wiring that is easiest to get wrong (`validate_workflow`)
- Templates searched by keyword, by the nodes they use, by task, or by metadata such as complexity and target audience, then fetched as nodes only, as structure, or whole (`search_templates`, `get_template`)
- Documentation for the tools themselves, which is where the maintainers suggest starting (`tools_documentation`)
- With API access configured: workflows created, listed, retrieved at four levels of detail, replaced wholesale or patched with diff operations, deleted, validated in place, auto-fixed, rolled back through version history, and deployed straight from an n8n.io template (`n8n_create_workflow`, `n8n_update_partial_workflow`, `n8n_autofix_workflow`, `n8n_workflow_versions`, `n8n_deploy_template`)
- Executions listed, read and deleted; a workflow triggered by webhook, form or chat; evaluation runs read and triggered; and folders, data tables and credentials managed (`n8n_executions`, `n8n_test_workflow`, `n8n_evaluations`, `n8n_manage_folders`, `n8n_manage_datatable`, `n8n_manage_credentials`)
- A security audit that combines n8n's own audit API with a deep scan of the workflows, and a connectivity check (`n8n_audit_instance`, `n8n_health_check`)
Nothing for the seven core tools: they take the workflow or the node config as an argument rather than reading it from a running n8n. The hosted service at dashboard.n8n-mcp.com needs only an API key and gives 100 tool calls a day on the free tier; self-hosting is documented for npx, Docker and Railway. The sixteen management tools need `N8N_API_URL` and `N8N_API_KEY` pointing at your instance, and an instance behind Cloudflare Access additionally needs `N8N_CF_CLIENT_ID` and `N8N_CF_CLIENT_SECRET`. For somewhere that changes should not be made by accident, `DISABLED_TOOLS` removes whole tools and `DISABLED_TOOL_OPERATIONS` removes individual destructive operations while keeping the reads.
One command plus a key — npx n8n-mcp, then supply credentials