n8n MCP

What does an n8n MCP server do?

That depends which one you mean. The instance-level server lets a client search, run, build and edit workflows across an instance. The MCP Server Trigger node exposes tools from one workflow. The MCP Client nodes run the other way, so n8n calls outside MCP servers. n8n-mcp is separate.

• Instance-level MCP server (maintained by n8n): one connection per instance, central sign-in, and workflows you enable one by one.
• MCP Server Trigger node (maintained by n8n): a single workflow acts as its own server, with its own URL and sign-in. You attach workflows with the Custom n8n Workflow Tool node.
• MCP Client Tool node (maintained by n8n): an n8n AI Agent calls tools on an outside server. The MCP Client node does the same as a normal workflow step.
• n8n-mcp (maintained by Romuald Członkowski, community): gives assistants n8n node documentation, templates and validation, plus workflow management if you add an n8n API key.
• n8n docs servers (from n8n, on GitBook and Kapa.ai): search n8n's docs, and on Kapa.ai its forum and blog too. They answer questions about n8n and don't read your instance.

Is there an official n8n MCP server?

Yes. n8n builds the instance-level server, and its blog says it ships in every edition: Cloud, Enterprise and the free self-hosted Community Edition. Anthropic's directory lists a connector made by n8n GmbH. n8n-mcp is a separate MIT-licensed project, and its privacy policy calls it independent and community-driven.

• n8n made its instance-level settings generally available in version 2.31.
• Romuald Członkowski operates the hosted n8n-mcp service through AiAdvisors in Poland.
• n8n's blog acknowledges community members who built their own open-source MCP servers, without naming a project.

What is the n8n MCP server URL, and how do you sign in?

Each instance has its own URL, so copy it from Settings > Instance-level MCP > Connect a client. It ends in /mcp-server/http. OAuth is n8n's recommended sign-in, and the API key tab gives a personal access token. MCP Server Trigger URLs are separate, with a test and a production URL per workflow.

• An instance owner or admin must enable MCP first, on Cloud or self-hosted.
• A new API token revokes the previous one, and API key clients don't appear in the OAuth revocation list.
• The MCP Server Trigger supports SSE and streamable HTTP, with no auth, Bearer auth, Header auth or n8n User Auth. It doesn't support stdio.
• n8n-mcp takes N8N_API_URL and N8N_API_KEY. Without them you get its documentation and validation tools only.
• n8n's docs servers are at https://docs.n8n.io/~gitbook/mcp and https://n8n.mcp.kapa.ai.

What can an n8n MCP server read and change?

The instance-level server searches workflows, runs the ones you enable and, from 2.13.0, creates and edits workflows and data tables. Clients can't read full workflow data, run or change a workflow until it's enabled for MCP. search_workflows still returns previews of every workflow the user can view.

What it can read

• Instance-level: search_workflows, get_workflow_details and get_workflow_history.
• Instance-level: search_workflow_executions and get_workflow_execution for run history.
• Instance-level: list_credentials lists credentials without secret data.
• n8n-mcp: search_nodes, get_node, search_templates and validate_workflow.

What it can change

• Instance-level: execute_workflow, test_workflow, publish_workflow and unpublish_workflow.
• Instance-level: create_workflow_from_code, update_workflow and archive_workflow, plus data table tools such as add_data_table_rows.
• Instance-level, agents in preview: call_agent runs real tools with real credentials, so it can change data and send messages.
• n8n-mcp with an API key: n8n_create_workflow, n8n_delete_workflow and n8n_manage_credentials.

How do you set up n8n MCP in Claude, ChatGPT and Cursor?

n8n names all three in its docs. An owner or admin enables MCP, you enable each workflow, then you open Connect a client and pick your client. From 2.33.0 that dialog shows the steps, and the client examples page covers older versions. Claude and ChatGPT connect from the cloud, so your instance must be publicly reachable.

{
  "mcpServers": {
    "n8n": {
      "type": "streamable-http",
      "url": "https://<your-n8n-domain>/mcp-server/http"
    }
  }
}

What can you ask an n8n MCP server?

Start with reads that name a tag or date range. These examples use tools n8n lists for the instance-level server, and full data only comes back from workflows enabled for MCP. The build prompt needs 2.15.0 or later, because it also tests. Leave new workflows unpublished until a person checks them, because publish_workflow activates them.

• Reads

  Search workflows tagged lead-routing and list each one's name, trigger and last update.

  Copy

• Reads

  List failed executions of our webinar sync workflow from the last 7 days. For each one, show which node failed.

  Copy

• Reads

  Show what changed between the last two versions of our lead scoring workflow.

  Copy

• Writes

  Run the published weekly report workflow, then tell me the execution ID and whether it succeeded.

  Copy

• Writes

  Build a draft workflow that adds form signups to a data table. Test it with sample data and leave it unpublished.

  Copy

What are the limits of an n8n MCP server?

Access isn't scoped per client, so every connected client sees all MCP-enabled workflows its user can reach. test_workflow times out at 300 seconds by default and 3,600 at most. search_workflows returns 200 results at most. n8n hasn't published a call limit for the server itself.

• execute_workflow returns an execution ID straight away, so the client checks the result with get_workflow_execution.
• From 2.47.0, a protected instance blocks every tool that isn't read-only, apart from execute_workflow in production mode.
• The MCP Server Trigger needs all /mcp* requests routed to one webhook replica if you run queue mode with several replicas.
• The n8n-mcp readme says never to edit production workflows directly with AI, and it documents a read-only deployment.
• Hosted n8n-mcp takes your n8n API key, which it says it encrypts with AES-256-GCM. Its telemetry is anonymized, and you can turn it off.

How much does an n8n MCP server cost?

The instance-level server comes with your n8n plan. The self-hosted Community Edition is free, and n8n's pricing page includes instance-level MCP on Starter (€20 a month billed annually, 2,500 executions), Pro (€50, 10,000 executions) and Enterprise. Hosted n8n-mcp is free for 100 tool calls a day, then €19 a month.

• n8n counts one execution per full workflow run, however many steps it has.
• Business is €667 a month billed annually for 40,000 executions, and it's self-hosted only. Enterprise pricing comes from n8n's sales team.
• The €19 n8n-mcp Supporter plan has a fair-use cap of 5,000 tool calls a day. Self-hosting it is free under the MIT licence.
• The Community Edition is free when only you or your organization build the workflows. Hosting n8n for clients to build in needs a different licence.

What are the alternatives to an n8n MCP server?

The main alternatives are another automation server, a fixed workflow with no assistant, or the app's own MCP server. Zapier MCP reaches 9,000+ apps with nothing to host. MCP vs API vs Zapier explains when a fixed workflow beats an assistant, and that logic applies to n8n workflows too.

• Zapier MCP: Hosted, with actions in 9,000+ apps through one connection.
• Workato MCP: Prebuilt app servers and servers built from your recipes.
• HubSpot MCP: HubSpot's own server, if it's the only app you need.
• A fixed n8n workflow: Known triggers and rules, with no assistant in the loop.

Which skill goes with an n8n MCP server?

Use the free Agent write safety skill before you let a client publish or run workflows, because a workflow can send messages and change records in every app it connects to. Use the Marketing context kit to write down your naming rules and definitions once, so the assistant has less to guess when it builds.

• Agent write safety: Sets approvals before an agent can publish or run a live workflow.
• Marketing context kit: Holds your naming rules and definitions in one folder for every prompt.