# How to set up the n8n MCP server > **Governing MCP across your organization?** > Speakeasy's [AI control plane](/product/ai-control-plane) connects, secures, and observes every agent and MCP server from one entry point — with SSO, RBAC, runtime guardrails, and a full audit trail. n8n has an official MCP server built into every n8n instance, on n8n Cloud and self-hosted. Its URL is your instance's address followed by `/mcp-server/http`. It lets an AI agent search your workflows, run them, and build new workflows for you. This guide shows you how to connect the n8n MCP server to Claude Code and Codex. You'll turn on MCP access in n8n, create an access token, choose which workflows your agent can use, and add the server to your agent. ## Prerequisites To follow along, you'll need: - An n8n instance on version 2.33 or later, on n8n Cloud or self-hosted. This guide uses n8n 2.44.5. - The owner or admin role on that instance, to turn on MCP access - [Claude Code](https://code.claude.com/docs) or [Codex CLI](https://developers.openai.com/codex) installed ## Step 1: Enable instance-level MCP in n8n In n8n, open **Settings** and select **Instance-level MCP**. Click **Enable MCP access**. ![n8n's Instance-level MCP settings page with the Enable MCP access button highlighted](/assets/mcp/using-mcp/n8n-mcp/enable-mcp-access.png) The page now shows **Connection details**, with **MCP status** set to **Enabled**. ## Step 2: Create an n8n MCP access token Under **Connection details**, click **Connect** next to **Connect your client**. In the **Connect a client** dialog, select the **API key** tab. n8n generates a personal access token the first time you open this tab, and it shows the token only once. Click the copy icon next to **Access token**. The token lets your agent act as you in n8n, so keep it out of config files and code. Instead, save it as an environment variable in your shell profile: ```bash echo 'export N8N_MCP_TOKEN="your-n8n-token"' >> ~/.zshrc ``` Replace `your-n8n-token` with the token you copied. If you use Bash, write to `~/.bashrc` instead. Open a new terminal window so the variable loads. Also copy the **Server URL** from the same tab. You need it in Step 4. When you come back to the tab later, n8n hides the token. To replace it, click the refresh icon next to the hidden value. This revokes the old token, so update your agents with the new one. ![The Connect a client dialog on the API key tab, with the Server URL and the hidden access token](/assets/mcp/using-mcp/n8n-mcp/api-key-tab.png) ## Step 3: Make workflows available in MCP Your agent can see a preview of every workflow you have access to, but it can only run a workflow that's available in MCP. To make a workflow available, go to your workflows list in n8n. Click the **...** menu on the workflow's card and select **Enable MCP access**. ![The n8n workflows list with the menu open on the Score inbound lead workflow and Enable MCP access highlighted](/assets/mcp/using-mcp/n8n-mcp/enable-mcp-access-workflow.png) The card now shows the MCP icon after its creation date. You can also turn on **Available in MCP** in a workflow's **Settings** inside the editor. Workflows that your agent creates through MCP are available in MCP automatically. ## Step 4: Connect your agent to the n8n MCP server Follow the instructions for the agent you use: [Claude Code](#claude-code) or [Codex](#codex). For any other agent, see [Connect other MCP clients](#connect-other-mcp-clients). In the commands below, replace `` with the start of the **Server URL** you copied in Step 2, for example `https://your-name.app.n8n.cloud`. For a local n8n instance, use `http://localhost:5678`. ### Claude Code Add the n8n MCP server to Claude Code: ```bash claude mcp add --transport http --scope user n8n /mcp-server/http --header 'Authorization: Bearer ${N8N_MCP_TOKEN}' ``` Keep the single quotes around the header. They tell Claude Code to save the variable name rather than the token itself, and Claude Code reads the token from your environment when it starts. The `--scope user` option makes the server available in all your projects. Check that the server connects: ```bash claude mcp list ``` The `n8n` server shows as **Connected**. On a local instance, the line looks like this: ```text n8n: http://localhost:5678/mcp-server/http (HTTP) - ✔ Connected ``` Start Claude Code with `claude` and ask it to run one of your workflows, for example: ```text Use my n8n Score inbound lead workflow to score this lead: Dana Lee from Northwind Logistics, 400 employees, a $45,000 budget, buying within 2 months ``` Replace the workflow name and details with one of your own workflows that's available in MCP. Claude Code asks for permission the first time it uses each n8n tool. Select **Yes** to let it run. Claude Code finds the workflow, runs it with your input, and reports the result. ![Claude Code running the Score inbound lead workflow and reporting a score of 72 out of 100 in the hot tier](/assets/mcp/using-mcp/n8n-mcp/claude-code-answer.png) ### Codex Add the n8n MCP server to Codex: ```bash codex mcp add n8n --url /mcp-server/http --bearer-token-env-var N8N_MCP_TOKEN ``` The `--bearer-token-env-var` option tells Codex to read the token from the `N8N_MCP_TOKEN` environment variable and send it with each request. Check the server's configuration: ```bash codex mcp list ``` On a local instance, the `n8n` row looks like this: ```text Name Url Bearer Token Env Var Status Auth n8n http://localhost:5678/mcp-server/http N8N_MCP_TOKEN enabled Bearer token ``` Start Codex with `codex` and enter `/mcp`. The `n8n` server shows as `connected (50 tools)`. Ask Codex to build a workflow, for example: ```text Create an n8n workflow that gets the weather forecast for London from the Open-Meteo API every morning at 7am ``` Codex asks for permission before it runs some n8n tools, such as `update_workflow` and `execute_workflow`. Select **Allow** to let it continue. Codex builds the workflow, tests it, and publishes it. ![Codex reporting that it created, tested, and activated the London Weather — Daily 7am workflow](/assets/mcp/using-mcp/n8n-mcp/codex-answer.png) The new workflow appears in n8n, published and ready to run. ![The London Weather — Daily 7am workflow in the n8n editor, with a schedule trigger and an HTTP Request node](/assets/mcp/using-mcp/n8n-mcp/codex-workflow-in-n8n.png) ## Connect other MCP clients You can connect any MCP client that supports remote (Streamable HTTP) servers and custom headers. Use these values: | Setting | Value | |---|---| | Server URL | `/mcp-server/http` | | Header | `Authorization: Bearer ` | ## Troubleshooting ### Codex shows `n8n: failed (0 tools)` If `/mcp` in Codex shows the n8n server as failed, press `F2` to view the warning: ```text MCP client for `n8n` failed to start: MCP startup failed: Environment variable N8N_MCP_TOKEN for MCP server 'n8n' is not set MCP startup incomplete (failed: n8n) ``` This warning means Codex started before the `N8N_MCP_TOKEN` variable was set. Codex runs a background process that keeps the environment it started with, so you need to restart it. Quit Codex, then stop its background process: ```bash pkill -f "codex app-server"; pkill -f codex-code-mode-host ``` Open a new terminal window, start `codex`, and enter `/mcp` again. ## Next steps You now have the n8n MCP server connected to your agent. You turned on instance-level MCP, created an access token, saved it as an environment variable, chose which workflows your agent can use, and added the server to Claude Code or Codex. Your agent can search your workflows, run them, and build new ones for you. To keep going: - Connect more services to your agent with the other guides in [Connect popular MCP servers](/mcp/using-mcp/mcp-server-providers). - Learn how to get better results from your agent's tools in [Best practices for using MCP tools](/mcp/using-mcp/using-tools). - See what teams build with MCP in [MCP use cases](/mcp/using-mcp/use-cases). --- Index of every Speakeasy page for agents: [llms.txt](https://www.speakeasy.com/llms.txt). AI control plane and MCP gateway content in one file: [llms-full.txt](https://www.speakeasy.com/llms-full.txt). Any page is available as markdown by appending `.md` to its URL or by requesting it with `Accept: text/markdown`.