Mission · Why this matters

Connect Claude Code to the platform — one command, the line is live

In the Agentowner course you gave your KI-Agent (AI agent) its Token. Today you use that Token yourself for the first time: a single command connects Claude Code to LinkYourSkill's MCP access, and you see with your own eyes what the platform offers your agent there. No chat involved yet — first the line, then the brain.

Lesson 01 · ~7 minutes · Knowledge & practice

What you'll be able to do

Connect the MCP access to Claude Code with a single command — using your Token.
Check which Tools your agent is offered before you ever chat — with /mcp.

You already steer — now you build

From the Agentowner course you know the rule: your KI-Agent acts in your name, and nothing becomes real before you approve it. So far the agent has been finished software to you. In this course you build it yourself — here in the terminal, with Claude Code. It's the terminal flavor of "Your agent in n8n": same through-line, same marketplace, only your tool is different. You don't need n8n here.

Core idea

The Token (your key) you created in the Agentowner course is exactly what your agent identifies itself with at this access point. There is no second secret: whoever holds the Token acts as your agent — inside your limits. That's why it stays as secret as a password, in the terminal too.

Before you start

You've done the Agentowner course: an agent exists, and its Token is stored safely with you.
Claude Code is installed and signed in. Quick test: type claude — the prompt appears. Anthropic describes the install at docs.claude.com/claude-code.
A terminal you're comfortable in. Verified with Claude Code 2.1.201.

This course is allowed to get technical

The Agentowner course deliberately stays at the "key" altitude — no jargon. This builder track goes one level deeper: you'll see terms like MCP, Bearer and real commands. Claude Code's commands stay in English, word for word — so you find them again in your terminal exactly as written.

What is the MCP access?

MCP (Model Context Protocol) is the standardized access an AI agent uses to talk to LinkYourSkill. Three things are enough to understand it:

One endpoint: https://app.linkyourskill.ai/mcp — an address your agent sends its requests to.
Your Token: it is sent along with every request as a "Bearer Token" — that's how your agent identifies itself.
Tools: the access offers a catalog of Tools — individual capabilities like search_skillanbieter. Which of them your agent may use is your decision, later.

The first connection — one command

In Claude Code you need no workflow and no building blocks. A single command connects the MCP access. Open your terminal and enter it — putting your own Token where marked (don't type the angle brackets):

claude mcp add --transport http lys https://app.linkyourskill.ai/mcp --header "Authorization: Bearer <your-token>" -s user

Claude Code confirms:

Added HTTP MCP server lys with URL: https://app.linkyourskill.ai/mcp to user config Headers: { "Authorization": "[REDACTED]" } File modified: ~/.claude.json

What's inside that command:

lys — the name you give the access. Free to choose; we keep lys, because your Tools will be named mcp__lys__… later.
--transport http — the MCP access speaks "HTTP Streamable". That's exactly the transport you specify.
--header "Authorization: Bearer <your-token>" — this is where your Token travels along.
-s user — the access applies across all your projects and is stored nowhere you might share by accident.

Token hygiene

In its confirmation Claude Code shows [REDACTED] instead of your Token — good. It is stored in cleartext in ~/.claude.json. Two things follow: claude mcp get lys shows the Token in full (so don't run it on a shared screen), and if you use -s project instead of -s user, it lands in a file .mcp.json in the project folder — handy for a team, but never commit it to Git. With -s project Claude Code also asks once to approve the server on first start (status ⏸ Pending approval). -s user avoids both.

See what's offered — `/mcp`

First the line, then the brain. Before you chat with your agent, you check the connection. Start Claude Code with claude and enter the command /mcp:

Manage MCP servers lys · ✔ connected

✔ connected means the line is live. Select lys to see the whole catalog — capabilities like search_skillanbieter, prepare_order and get_order_status. Which of them your agent may actually use is your call — we get to that in Lesson 4.

Your deterministic proof

/mcp shows the catalog without the AI doing anything. No chat, no guessing: the connection is live before a single word reaches the model. That's exactly what "first the line, then the brain" means.

If it doesn't work

⏸ Pending approval — you used -s project. Start claude and approve the server once. With -s user this doesn't happen.
No ✔ connected, an authentication error instead — the Token is missing or wrong. Check it with claude mcp get lys (shows it in cleartext); if in doubt, create a new Token for your agent in the Agentowner dashboard.
Just to experiment without your own Token: the platform's MCP docs name a public demo key that can only read (no prepare_order).

Your turn

Test yourself — retrieving from memory is what builds real retention.

Knowledge check

How do you check the line is live — before the AI does anything?

/mcp shows the tool catalog and the connection status without the AI doing anything — your deterministic proof that the line is live. A test message would already engage the model (and call Tools), and the web page says nothing about your agent's connection.