TypingMind MCP + n8n

Connect n8n to TypingMind with n8n’s built-in MCP server. Build, test, and publish n8n workflows by chatting, and turn your own workflows into tools the AI can call.

Connect TypingMind to n8n and let the AI do the automation work. Describe a workflow in plain English, and the AI builds it in your n8n, tests it, fixes what breaks, and publishes it. You can also turn your own n8n workflows into tools that the AI calls during a chat.

This guide works with n8n Cloud, self-hosted n8n on a public HTTPS URL, and n8n on your computer (localhost). The TypingMind steps are the same; only the Server URL and which connector you pick change.

The AI doesn’t only build workflows. You can also run your n8n workflows right from a TypingMind chat and get the results back in the chat. See Run a workflow from chat.

This guide walks through two real examples:

1. Build a support ticket workflow by chatting. It sorts incoming tickets by priority and category, saves them, replies to the sender, and sends a digest every day at 5 PM.
2. Turn a workflow into AI tools. It’s a mini CRM that you can talk to: “Add Jane from Acme”, “Who’s still new?”

This guide was tested with n8n 2.46.4 (self-hosted), TypingMind MCP Connector 1.4.1, and Claude Haiku 5.5 on November 4–5, 2026. n8n’s built-in MCP server is free in all n8n editions, including self-hosted Community Edition.

Ways to connect n8n

There are three ways to connect n8n to TypingMind. You can use more than one at the same time.

	n8n’s built-in MCP server (instance-level MCP)	MCP Server Trigger node	n8n-mcp (third party)
What the AI can do	Build, edit, test, run, and publish workflows. Read and write data tables.	Call only the tools that one workflow exposes.	Look up n8n node docs, validate workflows, and manage them through the n8n API.
Best for	Building and managing automations by chatting	Giving the AI a few safe, specific actions (your “custom tools”)	Users who already use it, or n8n versions without the built-in server
Server URL	https://<your-n8n>/mcp-server/http	https://<your-n8n>/mcp/<path>	Runs with npx, or as a hosted service
Auth	n8n MCP access token	None, Bearer token, or header auth (you choose)	n8n API key
Cost	Free	Free	Free to self-host, paid hosted plan
In this guide	Part 1 and Part 2	Part 3	Other option

Requirements

• An n8n instance in any of these forms:
  • n8n Cloud
  • Self-hosted n8n with a public HTTPS URL
  • Local n8n on localhost (or another private address)
• You need to be the owner or an admin to turn on MCP access.
• An AI model that supports plugins/tool calling. Claude Haiku 5.5 handled every step in this guide.
• Connector:
  • n8n Cloud or public HTTPS: TypingMind Cloud is enough. You don’t need the Private MCP Connector.
  • Local / private network: the Private MCP Connector running on a computer that can reach n8n, usually the same computer.

Part 1: Connect n8n to TypingMind

Pick the tab that matches your n8n. n8n Cloud or a public HTTPS URL is the simpler path — TypingMind Cloud talks to n8n directly, and you don’t need the Private MCP Connector. Use the localhost tab if n8n runs on your computer.

n8n Cloud or public HTTPS URL

1. Turn on MCP access in n8n

   1. In n8n, click Settings (bottom left) and choose Instance-level MCP.

   2. Under Connection details, check that MCP status says Enabled. If it doesn’t, open the menu and enable it.

   3. Under Access, Workflows enabled shows which workflows MCP clients can use (for example, 2 workflows).

      
2. Copy the Server URL and access token

   1. Next to Connect your client, click Connect.

   2. In the Connect a client dialog, open the API key tab.

   3. Copy the Server URL. On n8n Cloud it looks like https://<your-instance>.app.n8n.cloud/mcp-server/http. On self-hosted n8n with a public domain, it looks like https://<your-domain>/mcp-server/http. The URL must use https and end with /mcp-server/http.

   4. Next to Access token, click the refresh icon to create a token (or copy the one you already generated), then click the copy icon next to it.

      

   n8n shows the full access token only once (“Make sure to copy your access token, you won’t be able to see it again”). After that, the field is masked (for example ******C0-Y) and you only get a Generate new token action. Save the token somewhere safe when you first create it. Generating a new token revokes the old one, so you’d have to update the header in TypingMind.

3. Add n8n as an MCP server in TypingMind

   1. Click Plugins in the left sidebar.

   2. Under Store, click MCPs, then click + Add MCP (top right).

      

   3. The Add Custom MCP Server URL page opens (MCPs / Add URL). Fill in:

      • Server URL: the HTTPS Server URL from n8n, for example https://your-n8n.example.com/mcp-server/http
      • Connection name: n8n (or n8n Cloud)
      • Description (optional): for example, “Build and run n8n workflows”.

   4. Open Advanced options, turn on Custom HTTP headers, and add:

      • Header name: Authorization
      • Header value: Bearer <your n8n MCP access token>

   5. Click Create connection.

      

   Type the word Bearer, a space, and then the token. If you forget Bearer , n8n rejects the connection with 401.

4. Choose TypingMind Cloud

   The new entry shows Connection Required via Select Connector. Click Connect. In the Select MCP Connector dialog, choose TypingMind Cloud, then click Select Connector. You don’t need the Private MCP Connector for this path.

   
5. Check the connection and tools

   When it works, the entry shows ✓ Connected via TypingMind Cloud.

   

   Open the Tools tab. You should see Tools (55), each with a toggle and a Read-only or Destructive badge.

   

   All 55 tool descriptions are sent to the model with every message. That’s about 21,000 tokens per request. To save tokens:

   • Turn off tools you don’t need in the Tools tab. For building workflows, the agent and skill tools (search_agents, create_agent, call_agent, search_n8n_skills, create_skill, and the rest) can usually be turned off. They make up about 40% of the total.
   • Turn on the n8n plugin only in chats where you need it.
6. Turn on n8n in a chat

   Start a new chat. Click the plugin selector next to the model name, type n8n in Search, and turn on n8n under MCP.

   

   Check that it works:

   List my n8n workflows.
   

   The AI calls n8n’s workflow search tool and lists your workflows, with their names, IDs, and what they do. (In the screenshot, the IDs are blurred.)

   

   The AI can only see workflows that are available in MCP. Workflows the AI creates are available automatically. To make an existing workflow available, open it in n8n, click … → Settings, and turn on Available in MCP.

n8n on your computer (localhost)

1. Start n8n

   If you already have n8n on localhost, skip this step.

   To try n8n on your computer, install Docker and run:

   docker run -it --rm --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8n
   

   Open http://localhost:5678 and create your owner account.

2. Turn on MCP access in n8n and copy your token

   1. In n8n, click Settings (bottom left) and choose Instance-level MCP.

   2. Under Connection details, check that MCP status says Enabled. If it doesn’t, open the menu and enable it.

   3. Under Access, Workflows enabled shows which workflows MCP clients can use (for example, 2 workflows).

      

   4. Next to Connect your client, click Connect.

   5. In the Connect a client dialog, open the API key tab.

   6. Copy the Server URL: http://localhost:5678/mcp-server/http.

   7. Next to Access token, click the refresh icon to create a token, then click the copy icon next to it.

      

   n8n shows the token only once (“Make sure to copy your access token, you won’t be able to see it again”). After that, the field shows only the last few characters. Save the token somewhere safe. Clicking the refresh icon again creates a new token and the old one stops working, so you’d have to update TypingMind.

   Use the API key tab for this guide. The Private MCP Connector doesn’t support OAuth sign-in.

3. Start the Private MCP Connector

   TypingMind can’t reach localhost by itself. The Private MCP Connector runs on your computer and talks to n8n for TypingMind, so you don’t need a tunnel or a public URL.

   In TypingMind, go to Settings → Advanced Settings → Model Context Protocol, click Setup Connector, choose This Device, and run the command it shows in a terminal. Keep the terminal open.

   npx @typingmind/mcp <your-token>
   

   For details, see Use MCP with Private MCP Connector. If it won’t connect, see Can’t connect to the MCP Connector?

4. Add n8n as an MCP server in TypingMind

   1. Click Plugins in the left sidebar.

   2. Under Store, click MCPs, then click + Add MCP (top right).

      

   3. The Add Custom MCP Server URL page opens (MCPs / Add URL). Fill in:

      • Server URL: http://localhost:5678/mcp-server/http
      • Connection name: n8n
      • Description (optional): for example, “Build and run n8n workflows”.

   4. Open Advanced options, turn on Custom HTTP headers, and add:

      • Header name: Authorization
      • Header value: Bearer <your n8n MCP access token>

   5. Click Create connection.

      

   Type the word Bearer, a space, and then the token. If you forget Bearer , n8n rejects the connection.

5. Choose Private MCP Connector

   The new n8n entry shows Connection Required via Select Connector. Click Connect. The Select MCP Connector dialog opens with two options, TypingMind Cloud and Private MCP Connector.

   Choose Private MCP Connector. When the connector is running, the dialog shows Connected to localhost:50881 (“Your MCP requests will be routed through your private connector”). Click Select Connector. To skip this dialog next time, tick Remember this choice for all future connections.

   

   When it works, the n8n entry shows ✓ Connected via Private MCP Connector.

   
6. Check the tools

   Open the Tools tab of the n8n entry. You should see Tools (55), each with a toggle and a Read-only or Destructive badge.

   

   All 55 tool descriptions are sent to the model with every message. That’s about 21,000 tokens per request. To save tokens:

   • Turn off tools you don’t need in the Tools tab. For building workflows, the agent and skill tools (search_agents, create_agent, call_agent, search_n8n_skills, create_skill, and the rest) can usually be turned off. They make up about 40% of the total.
   • Turn on the n8n plugin only in chats where you need it.
7. Turn on n8n in a chat

   Start a new chat. Click the plugin selector next to the model name, type n8n in Search, and turn on n8n under MCP.

   

   Check that it works:

   List my n8n workflows.
   

   The AI calls n8n’s workflow search tool and lists your workflows, with their names, IDs, and what they do. (In the screenshot, the IDs are blurred.)

   

   The AI can only see workflows that are available in MCP. Workflows the AI creates are available automatically. To make an existing workflow available, open it in n8n, click … → Settings, and turn on Available in MCP.

Part 2: Build a support ticket workflow by chatting

In this example, a support form sends tickets to n8n. The AI builds a workflow that:

• Receives each ticket (name, email, message) through a webhook.
• Sorts the ticket by priority (urgent, normal, or spam) and category (billing, technical, feature request, or general).
• Saves urgent tickets as needs_reply_now and normal tickets as queued in an n8n data table called support_tickets.
• Replies to the form with how long the customer will wait: 1 hour for urgent tickets, 1 business day for normal ones. Spam is ignored.
• Every day at 5 PM, builds a digest of the queued tickets, grouped by category.

You don’t need any other accounts. The workflow uses only built-in n8n nodes.

1. Ask the AI to build the workflow

   With the n8n plugin turned on, send:

   Create an n8n workflow for my support form: a webhook receives name, email, message. Classify each ticket as urgent, normal or spam and by category (billing, technical, feature request). Save urgent and normal tickets in a data table called support_tickets with a status, reply to the form with an ETA, and ignore spam. Every day at 5 PM, build a digest of queued tickets. Use the webhook path support-intake. Then test it with 3 sample tickets.
   

   The AI works in several steps. It reads n8n’s workflow reference, looks up the nodes it needs, creates the support_tickets data table, validates the workflow, and then saves it to n8n.

   

   Building takes a few minutes and many tool calls. If validation fails, n8n explains what’s wrong, and the AI fixes it and tries again on its own. That’s normal.

2. See the workflow in n8n

   Open n8n in another tab. The new workflow is in your Overview list. Open it to see the canvas:

   

   Your node names may differ a little, but the shape is the same: Webhook → Classify Ticket → Route by Priority, with three branches (urgent, normal, spam), plus a separate Every Day 5PM → Get Queued Tickets → Build Daily Digest path.

   The AI sorts tickets with keyword rules in a Code node (for example, “refund”, “outage”, or “can’t log in” means urgent). An AI node in n8n would need an LLM API key saved in n8n, which the AI can’t create for you. See Limits.

3. Test it with sample tickets

   If you included “test it” in your first message, the AI runs the test right away. You can also ask for it:

   Test the workflow with these tickets:
   1. Jane, [e-mail]: "I was charged twice this month, need a refund ASAP"
   2. Tom, [e-mail]: "Feature idea: dark mode would be nice"
   3. Max, [e-mail]: "Cheap SEO services, free money!!!"
   Show me how each one was classified and what was saved.
   

   The AI runs the workflow once per ticket (execute_workflow) and reads each result (get_workflow_execution).

   

   In n8n, open Executions to see each run, and open the support_tickets data table to see the saved rows. The spam ticket isn’t saved.

   

   Test runs are real runs. They write real rows to your data table, and in your own workflows they can send real emails or messages. Use test data, and clear the table before you go live.

4. Ask the AI to improve it

   Now make the workflow safer with a follow-up message:

   Add error handling: if email or message is missing, reply with a 400 error instead of saving. Also retry saving urgent tickets up to 3 times.
   

   The AI changes only what’s needed (update_workflow). It adds a Valid Submission? check after the webhook that replies with 400 and {"ok":false,"error":"email and message are required"}, and turns on Retry On Fail (3 tries) for the node that saves urgent tickets.

   

   Then ask it to check the change:

   Test it with a ticket that has no email.
   

   More ideas to try:

   • “Add a separate branch for billing tickets.”
   • “Also mark tickets that mention ‘cancel’ as urgent.”
5. Publish the workflow

   A workflow only listens on its real (production) URL and runs on schedule after it’s published. Ask:

   Publish the workflow and give me the production webhook URL.
   

   The AI calls publish_workflow. The production URL looks like http://localhost:5678/webhook/support-intake.

   Point your form at this URL, or send a test request from a terminal:

   curl -X POST http://localhost:5678/webhook/support-intake \
     -H "Content-Type: application/json" \
     -d '{"name":"Ann","email":"[e-mail]","message":"App is down, cannot log in"}'
   

   The response:

   {"ok":true,"priority":"urgent","message":"We are on it. Expect a reply within 1 hour."}
   

   

   The 5 PM digest uses your n8n instance’s time zone. To change it, open the workflow’s Settings in n8n and set Timezone.

Run a workflow from chat

You can run any of your n8n workflows from a TypingMind chat, not only the ones the AI built. Ask in plain English, and the AI starts the workflow in n8n, waits for it to finish, and reads the result back into the chat.

Example prompts:

Run the Support Triage workflow with this ticket: Ann, [e-mail], "App is down, cannot log in". Tell me how it was classified.

Run the published version of Support Triage with the same ticket.

Show me the last 5 runs of Support Triage. Did any of them fail, and why?

What happens:

1. The AI starts the workflow with execute_workflow. The workflow runs on your n8n instance, not in TypingMind.
2. n8n returns an execution ID, and the AI reads the result with get_workflow_execution.
3. The AI answers in the chat with what each step did. For past runs, it uses search_workflow_executions.

To see every step in detail, open Executions in n8n. The test runs in Part 2 work the same way:

Good to know:

• Existing workflows need Available in MCP. Open the workflow in n8n, click … → Settings, and turn on Available in MCP. Workflows the AI creates through MCP have it turned on already.
• Draft or published version. By default the AI runs the current draft (test mode). Say “run the published version” to run the version that’s live on your production URL. The workflow must be published first.
• Workflows that need input. For workflows that start with a Webhook, Chat, or Form trigger, give the input data in your message (for example, the ticket fields). If a workflow has more than one trigger, tell the AI which one to use.
• Runs are real. Every node really runs. Data gets written, and emails or messages get sent.
• No waiting steps. Workflows that pause for a person, such as multi-step forms or approval (Wait) steps, can’t be run from chat.

Part 3: Turn a workflow into AI tools (Mini CRM)

Part 2 lets the AI build workflows. You can also go the other way: build a workflow once, and let the AI use it as a set of tools. n8n’s MCP Server Trigger node turns a workflow into its own MCP server.

In this example, the tools work on a small CRM stored in an n8n data table called leads:

Tool	What it does
add_lead	Adds a lead (name, company, email, note) with stage new
find_leads_by_stage	Lists leads in a stage: new, contacted, won, or lost
update_lead_stage	Moves a lead (matched by email) to a new stage

1. Build the Mini CRM workflow

   You can build it in n8n by hand, or ask the AI with the n8n plugin turned on:

   Create an MCP server workflow called Mini CRM with tools to add a lead, list leads by stage, and move a lead to a new stage, stored in an n8n data table called leads. Use the path mini-crm. Publish it and give me its MCP URL.
   

   The AI creates a workflow with an MCP Server Trigger and three Data table tools, publishes it, and gives you the URL. It looks like http://localhost:5678/mcp/mini-crm.

   
2. Add it to TypingMind

   Add a second MCP server the same way as in Part 1: Plugins → Store → MCPs → + Add MCP.

   • Server URL: http://localhost:5678/mcp/mini-crm (use the Production URL from the MCP Server Trigger node)
   • Connection name: mini-crm

   Choose Private MCP Connector for a localhost URL. The entry shows Tools (3): add_lead, find_leads_by_stage, and update_lead_stage.

   

   With Authentication set to None, anyone who can reach your n8n can call these tools. If your n8n is reachable from the internet, open the MCP Server Trigger node in n8n, set Authentication to Bearer Auth, and create a credential with a secret token. Then, in TypingMind, add the header Authorization: Bearer <that token> under Advanced options → Custom HTTP headers.

3. Talk to your CRM

   Start a chat, turn on mini-crm in the plugin selector, and try:

   Add Jane Doe from Acme, [e-mail], met at the webinar, wants the team plan.
   

   Add Bob Tran from Globex, [e-mail], asked for pricing.
   

   Move Bob to contacted.
   

   Who's still new?
   

   

   Only these three tools are sent to the model, so this uses far fewer tokens than the full n8n plugin. It’s also safer: the AI can only do what the workflow allows.

   To add more tools later, ask the AI with the n8n plugin: “Add a tool to delete a lead and one that counts leads per stage.” TypingMind picks up the new tools from the same URL.

Other option: the third-party n8n-mcp server

n8n-mcp is a community project by czlonkowski, not made by n8n or TypingMind. It gives the AI detailed docs for every n8n node and workflow templates, and it can manage workflows through the n8n API. It’s a good choice if you already use it, or if your n8n version doesn’t have the built-in MCP server.

To run it locally with the Private MCP Connector, go to Settings → Advanced Settings → Model Context Protocol → Edit Servers and add:

{
  "mcpServers": {
    "n8n-mcp": {
      "command": "npx",
      "args": ["n8n-mcp"],
      "env": {
        "MCP_MODE": "stdio",
        "LOG_LEVEL": "error",
        "DISABLE_CONSOLE_OUTPUT": "true",
        "N8N_API_URL": "https://your-n8n-instance.com",
        "N8N_API_KEY": "your-n8n-api-key"
      }
    }
  }
}

Create the n8n API key in n8n under Settings → n8n API → Create an API key. A hosted version is also available at n8n-mcp.com. Follow their setup guide to get the server URL and key.

Limits

• The AI can’t create credentials. It can’t add API keys or log in to apps (Gmail, Slack, OpenAI, and so on) for you. Add credentials in n8n yourself, then ask the AI to use them.
• AI nodes need an LLM credential. n8n’s AI Agent and LLM nodes need an API key saved in n8n. That’s why the example sorts tickets with keyword rules instead of an AI node.
• Test runs are real. When the AI tests a workflow, every node really runs. Data gets written, and messages get sent.
• No waiting steps in tests. The AI can’t test workflows that pause for a person, such as multi-step forms or approval steps.
• Only workflows available in MCP. The AI can’t see or run your existing workflows until you turn on Available in MCP for them.
• Building uses a lot of tokens. Expect several minutes and many tool calls for a build. Choose a model with good tool calling.

Security notes

• The n8n MCP access token gives the AI the same access as your n8n account. Don’t share it, and rotate it in Settings → Instance-level MCP if it leaks.
• Tools marked Destructive (for example archive_workflow and delete_data_table_column) can change or remove things. Turn off the ones you don’t need in the Tools tab.
• Your AI model provider receives the workflow details and data that the AI reads, the same as any chat.

Troubleshooting

TypingMind can't connect to n8n

• Check that the Server URL ends in /mcp-server/http. This isn’t the address in your browser’s address bar.
• Check the header: name Authorization, value Bearer followed by your token.
• Make sure MCP status says Enabled in Settings → Instance-level MCP.
• For a localhost URL, choose Private MCP Connector and make sure the connector is running. See Can’t connect to the MCP Connector?
• For n8n Cloud or a public HTTPS URL, choose TypingMind Cloud. TypingMind Cloud can’t reach localhost.

Cloud path: 401, Failed to fetch, or connection fails

• 401 Unauthorized: the access token is wrong, was regenerated (the old one is revoked), or the header is missing the Bearer prefix. Generate a new token in n8n (Settings → Instance-level MCP → Connect → API key), copy it once, and update Authorization in TypingMind to Bearer <new token>.
• Server URL: must be https://…/mcp-server/http (https, and the path must end with /mcp-server/http). A bare n8n editor URL won’t work.
• MCP access disabled: in n8n, open Settings → Instance-level MCP and make sure MCP status says Enabled.

The AI doesn't see my existing workflows

Turn on Available in MCP for the workflow: open it in n8n, click … → Settings. You can also turn on Auto-expose new workflows in Settings → Instance-level MCP.

The webhook URL returns 404

The production URL only works after the workflow is published. Ask the AI to publish it, or click Publish in n8n. While testing in the n8n editor, use the test URL (/webhook-test/...) instead.

The AI keeps adding node groups or mentions a canvas warning

n8n returns a canvas layout warning on every save. It’s harmless. Tell the AI to ignore it.

My chat is slow or expensive

The full n8n plugin sends about 21,000 tokens of tool descriptions with every message. Turn off tools you don’t use in the Tools tab, turn the plugin off in chats that don’t need it, or use an MCP Server Trigger workflow like the Mini CRM for everyday tasks.

Can I run my workflows from TypingMind, or only build them?

You can run them too. Ask the AI to run a workflow, with input data if it needs some, and it reads the result back into the chat. Existing workflows need Available in MCP turned on. See Run a workflow from chat.