Tutorial · 6 of 8

Install skills in n8n

n8n is open-source workflow automation — the no-code/low-code platform devs actually like. Pair it with iClaude skills and you get the entire AI skill catalog as drag-and-drop nodes.

7 min read

How MCP works in n8n

n8n added a native MCP Client node in version 1.74. You point it at a server URL (or a command for local servers) and it exposes every tool the skill offers as an action you can drag into your workflow.

Two patterns to choose between:

Skill-per-workflow — drop the MCP Client into one workflow that needs it. Fast, isolated.
Webhook bridge — wrap a skill behind a Webhook trigger so anything (Zapier, Make, your app, another AI) can call it via HTTP.

Pattern 1: skill in a workflow

Pick a skill and copy its install line

On iclaude.io/skills filter by Stack → n8n and pick something useful (e.g. Postgres query, web scrape, OpenAI image gen). The install box gives you a one-liner:

terminal

$ npx iclaude install postgres-query --stack n8n

Run it on the machine where n8n lives

If n8n runs on your laptop, run it there. If it runs in Docker, run it inside the container:

terminal

$ docker exec -it n8n npx iclaude install postgres-query --stack n8n

The installer writes server config to a folder n8n watches: ~/.n8n/mcp/postgres-query/.

Restart n8n

n8n only scans MCP config at startup. Stop and start it.

terminal

# Local
# Press Ctrl+C in the n8n terminal, then:
$ npx n8n

# Docker
$ docker restart n8n

Add the MCP Client node

Open any workflow in n8n. Click the + to add a node. Search for MCP Client and select it.

In the node settings:

Connection type: Local server
Server: pick postgres-query from the dropdown (it appears because of step 2)
Tool: pick the action you want, e.g. execute_query

Add credentials and test

Most skills need credentials — a database URL, an API key, etc. n8n stores these in its credential vault. Click Create New Credential in the node, fill in the values, hit save.

Click Execute Step. The node runs the skill once and shows the result. Green = working.

Pattern 2: webhook bridge

Want to call a skill from outside n8n? Wrap it behind a Webhook trigger. This turns any iClaude skill into a normal HTTPS endpoint.

Create a new workflow

New workflow → add a Webhook trigger node. Set the HTTP method (usually POST) and copy the production URL it generates — you'll use this URL to call the skill.

Wire the MCP Client to the webhook

Add an MCP Client node after the Webhook. Map the webhook body fields onto the skill's inputs using expressions, e.g. {{ $json.query }}.

Return the result

Add a Respond to Webhook node at the end. Set its body to {{ $json }} so the caller gets the skill's output. Save and activate the workflow.

Test from anywhere

terminal

$ curl -X POST https://your-n8n.example.com/webhook/postgres-query \
$   -H "Content-Type: application/json" \
$   -d '{"query":"SELECT count(*) FROM customers"}'

You're now exposing an iClaude skill as a normal HTTP API. Any tool that can call a URL can use it.

Common gotchas

MCP Client node missing. Update n8n. The node shipped in 1.74; older versions don't have it.
Dropdown shows no servers. The install ran outside n8n's home folder. Re-run with --n8n-home ~/.n8n (or wherever your install lives).
Skill works on first run, fails on second. Many MCP servers expect stateless calls; n8n by default reuses the server process. In the node settings, set Process Mode → Per Execution.
Docker can't reach the skill. Local skills installed on the host machine aren't visible inside the container — install inside the container (step 2) or mount the MCP folder as a volume.