Short answer: rent a small Linux server for roughly $5–10 a month, run n8n on it with Docker, put it behind your own domain with HTTPS, and switch on n8n’s built-in MCP access. You get unlimited workflows with no per-task bill, and Claude can call the workflows you choose as tools. Expect one to two hours for the first setup and a few minutes a month for updates.
Why bother self-hosting
Cloud automation tools charge per task, credit or execution. That is fine until an AI assistant starts calling your workflows dozens of times a day. Then the bill grows with every chat.
A self-hosted n8n turns that into a fixed cost. The server costs the same whether a workflow runs ten times or ten thousand. For lifetime-deal owners this fits the whole idea: pay once (or pay little), use a lot.
The trade-off is honest work: you look after a server. If you never want to see a terminal, the Make bridge is the easier route.
What it costs
- n8n itself: free. The Community edition is free to self-host under the Sustainable Use License, as long as you use it for yourself or your own organisation. Registering your email unlocks a few extra features, still free.
- A server: about $5–10 a month. A 2 vCPU / 4 GB machine is a good start. At Hetzner that is the CX23, which costs around €5–6 a month including an IPv4 address after the 2026 price rises, when it is in stock. Other providers sell similar machines in the same range.
- A domain: about $10–15 a year. If you already have one, a subdomain like
n8n.yourdomain.comcosts nothing extra. - Your AI usage: separate. If your workflows call OpenAI or Anthropic, you pay them directly for tokens. See our guide to bring-your-own-key tools.
Step by step
1. Rent the server. Choose Ubuntu as the system and the region closest to you. Add your SSH key during setup so you do not log in with a password.
2. Install Docker. Docker runs n8n in its own container, so updates are one command and nothing clashes with the rest of the system.
3. Write the Compose file. Follow n8n’s own Docker Compose guide. Two choices matter here:
- Use Postgres, not the default SQLite, if you expect more than light use. It handles many runs and large histories better.
- Set and save an encryption key. n8n uses it to encrypt your saved credentials. Lose it and every saved login becomes unreadable. Store it in your password manager.
4. Add a domain and HTTPS. Point n8n.yourdomain.com at the server and put a small reverse proxy such as Caddy in front of n8n. Caddy gets and renews the HTTPS certificate on its own.
5. Start it and create the owner account. Run docker compose up -d, open your domain and create the first user. Use a long, unique password.
6. Turn on MCP. In n8n, open Settings → Instance-level MCP and enable MCP access. Then open each workflow you want the AI to use and enable it for MCP. Nothing is exposed until you do both.
7. Connect Claude. Copy the server URL from n8n’s connection details (it ends in /mcp-server/http). In Claude, add a custom connector with that URL and approve the sign-in in n8n.
Two ways n8n speaks MCP
Instance-level MCP is the one above. One connection for the whole n8n server, one sign-in, and you pick which workflows are visible. Simplest for most people.
The MCP Server Trigger node turns one specific workflow into its own MCP server with its own URL and tools. Useful when you want separate, tightly scoped servers, for example one read-only and one that can write.
Whichever you use, keep the rule from the Make guide: expose small, specific workflows with clear names, not one workflow that can do everything.
Keeping it healthy
- Update monthly. Pull the new image and restart. n8n moves fast, and updates include security fixes.
- Back up. Turn on your provider’s server backups or snapshots and keep a copy of the encryption key elsewhere.
- Watch memory. If workflows start failing for no clear reason, check RAM first. Moving from 2 GB to 4 GB solves most problems.
- Limit exposure. Only the n8n domain should be reachable from the internet. Keep the database private.
When not to self-host
If you run two or three automations a week, Make’s free plan is simpler. If n8n would hold customer data and nobody on your team is comfortable with servers, pay for managed hosting. And if the tool you want to connect already has a working MCP server on your tier, connect it directly. A bridge is only worth it when it solves a real gap.