Labs MCP · Model Context Protocol
Labs MCP connects AI tools to iximiuz Labs, where they can coach you through hands-on learning, build custom playgrounds from plain-text prompts, and give coding agents real Linux servers for real server-side work.
https://labs.iximiuz.com/mcpWorks with Claude, ChatGPT, Codex, Cursor, opencode — any MCP client. And you decide exactly what it's allowed to do. Connection instructions ↓
› Spin up a Docker playground and run nginx on port 80.
✓start_play(docker) — docker-01 ready in 3s
✓run_shell_command(docker run -d -p 80:80 nginx)
✓scan_ports(docker-01) — HTTP · port 80
✓expose_port(80 · public)
↳ https://…-a1b2c3.iximiuz.com
nginx is live — open the URL above. Want me to spin up another container or clone a GitHub repo?
›
Four things it does
chapter 01 · learn
Tell your assistant what you're trying to learn. It searches the Labs catalog, assembles a plan that fits your background and your schedule — and when you get stuck, it acts like an infinitely patient mentor: right there in the same VM, nudging you forward with hints, not answers.
Your assistant can search the entire Labs catalog — tutorials, challenges, courses, skill paths. Tell it what you're after, and it'll pick the materials that match your level.
› Find me hands-on material about container networking — I already know the Docker basics.
There's no one-size-fits-all approach to learning. Tell your assistant where you are and where you want to get, and it'll assemble a personal learning path from Labs materials and pace you through it.
› Build me a two-week plan to get comfortable with Kubernetes — one hour a day, hands-on only.
Stuck mid-challenge, or a tutorial step isn't clicking? Your coach checks your task state, runs diagnostics in the same VM, and nudges you toward the next step — hints first, never the full solution.
› I'm stuck on the NAT challenge — the container pings the host but nothing outside. Hint please.
Progress is recorded server-side — streaks, completions, and daily practice stay in sync whether you learn in the browser or through your assistant.
chapter 02 · practice
Preparing a practice environment used to be a chore: writing init scripts, baking rootfs images, debugging the boot. Now it's a sentence: name the machines, the software, even the failure you want to debug — and your agent assembles the playground for you.
One VM, Docker preinstalled, a sample app already running behind a reverse proxy — ready for you to explore, break, and rebuild.
› Give me a Docker host with a small web app behind nginx — I want to practice tweaking reverse proxies.
A real K3s cluster with the tooling you asked for installed and wired up before you even open the terminal.
› Start the K3s playground, helm-install Argo CD, and give me the UI URL and the admin password.
Nothing teaches better than fixing a broken system. Ask your agent to sabotage a deployment, then debug it yourself — and let the agent verify your fix.
› Deploy a demo microservice app on K0s and break it in three subtle ways. Don't tell me what you broke.
Liked a setup? Ask your agent to save it as a reusable playground and start it again with one call.
chapter 03 · build
Your agent can stay right where it is — in a coding tool, a chat assistant, CI, or its own sandbox. When it needs somewhere real to run things, Labs MCP gives it up to 5 remote Linux machines wired into any topology — to build on, deploy to, test against, and leave behind in a reproducible state.
One script, many Linux distros. Let the agent run the same code against Ubuntu, Fedora, Debian, Rocky, Alpine, NixOS, and other real distro environments instead of guessing whether a fix is portable.
› This script works on Ubuntu but fails on Fedora. Test it across the mainstream Linux playgrounds and make it portable.
Sometimes the agent has a tiny container, a constrained runtime, the wrong kernel — or no execution environment at all. Give it a full Linux VM with the resources and capabilities the job actually needs.
› Build and run this application server on a fresh Linux VM, then expose it with a public URL so I can test it.
A server-side project is often more than one process. Give the agent several VMs, multiple networks, a Kubernetes cluster, external clients, databases, or whatever topology the system needs.
› Deploy this app to a K3s cluster, add the worker jobs and database, and test the service from a separate client VM.
Git worktrees isolate source trees, not CPUs, ports, databases, or dev servers. Let different branches — or different agents — use separate remote environments instead of competing for the same laptop.
› Deploy these three feature branches independently and give me a URL for each one.
Clone a repository, check out the exact commit, reproduce the failure, and preserve the resulting machine state. The repro can live beyond the agent session and be reopened later instead of reduced to a list of reproduction steps.
› Check out commit 8ac21f, reproduce issue #423, and save the working repro as a reusable playground.
chapter 04 · create PRO
iximiuz Labs isn't just for learners — authors publish hands-on tutorials and challenges here, and trainers run whole courses and workshops. Labs MCP brings an assistant into that workflow, too.
Publishing on iximiuz Labs? Draft tutorials, challenges, courses, and blog posts straight from your AI tool — the author tools work against your real author profile, and drafts land in your usual review flow.
› Draft a challenge where the student fixes a crashing systemd service — write the description and tasks, I'll review.
Running trainings or cohorts on Labs? Instructor assistance — course scaffolding, per-student environments, progress digests — is on its way.
Training on iximiuz Labs →Connect in a minute
Labs MCP is a standard remote MCP server (Streamable HTTP) — if your AI tool speaks MCP, it can connect. Two ways to authenticate it:
Add the URL and your client takes you through a browser sign-in — it registers itself and gets its own credentials, so there's nothing to copy around. Pick your client:
First protected action shows a consent screen listing exactly what's requested. Revoke anytime under Account → Connected apps.
For sandboxes, remote VMs, and CI — and for clients that don't speak OAuth.
An OAuth sign-in can end up redirecting to http://localhost:<port>/callback, where your client is expected to listen. That breaks on a remote VM or container (the browser redirects to the "wrong" localhost) and in CI or other headless environments (there may be no browser at all). As a workaround, you can mint a token once and pass it to the client as a plain header.
Get a token
in the browser
Account → Connected apps → Create access token. Choose the scopes it should get, then copy it.
in a terminal
labctl auth login # prints a URL — open it on any device labctl auth token
Use it
terminal
claude mcp add --transport http ixlabs https://labs.iximiuz.com/mcp --header "Authorization: Bearer <token>"
Other clients: add an Authorization: Bearer <token> header to the server config — check your client's docs for the exact key.
Permissions
Commands run only in playgrounds you own — your own sandbox — and every token is revocable.
accountauthorlearningplaygroundThe toolbox
Everything above boils down to plain MCP tools. Expand to browse the full list — with example prompts for each.
Open access
› Find beginner-friendly Kubernetes challenges on iximiuz Labs.
› Is there anything hands-on about eBPF on Labs?
› Which official networking challenges haven't I solved yet?
› What's the 'Reproduce a Docker Bridge Network' challenge about — and how hard is it?
› Which official Kubernetes playgrounds are there?
› List my custom playgrounds.
› How many nodes does the k3s playground have, and what's preinstalled?
› How does playground persistence work? Check the Labs docs.
› Read the Labs docs on custom playgrounds and walk me through making one.
› What topics does Labs cover around containers? Show me the tags.
› Who writes the Kubernetes material on Labs? Find me more from that author.
Read learning:read
› I'm on task 2 of the NAT challenge and iptables looks right to me — what am I missing?
› Explain what this task actually wants from me.
Write learning:write
› Start the 'Set Up NAT for a Container' challenge.
› Enroll me in the Docker The Hands-On Way roadmap.
› I've finished the bridge networking tutorial — mark it done.
Read playground:read
› I need kernel 6.1 for eBPF stuff — which kernels are available?
› Any RHEL-like base images — Rocky or Alma?
› Do I still have that K3s playground from yesterday?
› Is my playground done booting? When does it expire?
› How far am I in this challenge — which tasks are still failing?
› What's listening on this playground VM?
› Who can start my custom playground right now?
Write playground:write
› Start the Docker playground.
› Give me a clean Ubuntu 24.04 VM.
› Install nginx and curl localhost to check it's up.
› Run kubectl get pods -A and tell me what's crashlooping.
› Drop a docker-compose.yml with the app and Redis into /root.
› Expose port 3000 so I can open the app on my phone.
› Give me a browser terminal to this VM — I want to poke around myself.
› Don't let this playground vanish when the timer runs out.
› I'm done for today — stop the playground but keep my work.
› Bring back the cluster I stopped on Friday.
› I don't need this scratch VM anymore — destroy it.
› Title this run 'issue #423 repro'.
› The demo is at 5pm — give this play two more hours.
› Make me a reusable playground: Ubuntu with Go, Docker, and my dotfiles.
› Add a second VM to my custom playground for the database.
› Delete my scratch custom playground.
› Snapshot this setup as a playground so my study group can start from it.
› Make my k3s playground public.
Read account:read
› What plan am I on?
› Did I keep my streak alive this week?
› What's my daily practice today?
Write account:write
› Today's challenge is too hard — reshuffle my daily practice.
› Switch my digest to weekly.
Read author:read
› Do I have an author profile yet?
› Who can see my DNS tutorial draft?
Write author:write
› Set me up with an author profile called Jane Doe.
› Update my author bio to mention I write about eBPF.
› Start a draft tutorial titled 'Debugging DNS Inside Containers'.
› Share my draft challenge with github user octocat.
one url · any mcp client
Add the connector URL to your AI tool and start building. Happy hacking!
https://labs.iximiuz.com/mcp