Antrieb

12HTTPregistry active

Summary

Gives Claude real VMs to test infrastructure code instead of hoping it works in production. Provision multi-node clusters with Ubuntu, CentOS, VyOS, OpenWrt, or OPNsense in seconds, run commands via exec, snapshot working configs with save, then delete when done. Supports custom network topologies with multiple NICs per node for testing routers and firewalls. No local Docker or containers. The LLM writes a bash script or firewall rule, runs it on actual VMs with root access and real networking, sees it fail, adjusts, and tries again until it works. Hosted remote server, so you just add the URL to your MCP config and start breaking things in disposable 10-minute clusters.

CodeRabbit

AI writes the code. CodeRabbit catches the slop.

Try For Free →

Keep your Mac awake

Keep your Mac awake while Claude Code and 40+ AI agents run. Sleeps when they're idle.

One time payment $9 →

Context.dev

Integrate web data into your AI product. One API to scrape website & brand data.

Get API Key Now →

Make your agent a DeFi expert

Agent, run crypto. Access onchain data & trade routes via 1inch.

Install now →

Make money from your Skills

On Capafy, your Skill runs online 24/7 as an agent product, and you get paid every time someone uses it.

Start earning →

AppSignal

Monitor with ease. Code with confidence.

Start Free Trial →

CodeRabbit

AI writes the code. CodeRabbit catches the slop.

Try For Free →

Keep your Mac awake

Keep your Mac awake while Claude Code and 40+ AI agents run. Sleeps when they're idle.

One time payment $9 →

Context.dev

Integrate web data into your AI product. One API to scrape website & brand data.

Get API Key Now →

Make your agent a DeFi expert

Agent, run crypto. Access onchain data & trade routes via 1inch.

Install now →

Make money from your Skills

On Capafy, your Skill runs online 24/7 as an agent product, and you get paid every time someone uses it.

Start earning →

AppSignal

Monitor with ease. Code with confidence.

Start Free Trial →

Antrieb MCP Server

Instant disposable clusters for LLM-generated networks and infra.

Say an LLM writes a VyOS NAT config, an OpenWrt firewall rule set, or a bash script to harden a CentOS cluster. Where do you run it before it touches your environment? Antrieb gives the LLM a disposable multi-node cluster, with real VMs and real networking, to try what it generates. Break it, quickly reprovision, try again. No cleanup.

Same OS and appliances your LLM-generated code will target in production: CentOS Stream, Ubuntu, Debian, Alma, Alpine, VyOS, OPNsense, SONiC, OpenWrt. Multi-network topologies with per-NIC assignment. Not a container, not a microVM, not some unknown Linux — the same kernels and packages your real fleet runs.

Root access, private networking, passwordless SSH between nodes. Ten minutes per cluster. Clean slate every time. Instant clusters that change the economics of being wrong for LLMs.

Quick Start

Antrieb is a remote MCP server. Nothing to install. Add it to your config and start provisioning.

Claude Web / ChatGPT

With web-based AI clients, add Antrieb as a custom connector.

In claude.ai: go to Settings → Connectors, click Add custom connector. Enter:

Name: Antrieb
URL: https://antrieb.sh/mcp

Click Add, then Connect and sign in. Back in your chat, try something like:

Create a 3-node Kubernetes cluster, with node1 running both as control-plane and worker, and node2/node3 as pure workers. Install nginx with 3 pods and pod anti-affinity.

Walkthrough video: https://youtu.be/8nts8ol-yeA

MCP Clients (Claude Desktop, Claude Code, Codex, custom clients)

Get a free API key at antrieb.sh/dash, then add this to your MCP client config (e.g. .mcp.json for Claude Desktop):

Claude Code

Put this JSON in the current folder in a file called .mcp.json. Claude Code will find it automatically.

{
  "mcpServers": {
    "antrieb": {
      "type": "http",
      "url": "https://antrieb.sh/mcp",
      "headers": {
        "Authorization": "Bearer ant_YOUR_API_KEY"
      }
    }
  }
}

Codex

Put this TOML content in ./.codex/config.toml. Then set the ANTRIEB_TOKEN environment variable with your API key before launching Codex.

 [mcp_servers.antrieb]
  url = "https://antrieb.sh/mcp"
  bearer_token_env_var = "ANTRIEB_TOKEN"

No local install, no dependencies, no Docker.

How It Works

Your LLM controls real VMs through 5 tools:

1. provision  →  Spin up a cluster (sub-second per VM, declare custom networks and NICs)
2. exec       →  Run commands on any node (cluster-scoped env vars auto-injected)
3. save       →  Snapshot a node as an image, or save a runbook
4. search     →  Discover images, runbooks, or active clusters
5. delete     →  Destroy a cluster, or decommission a saved artifact

The LLM drives the entire workflow: provision a cluster, install software command by command, verify each step, and iterate until it works. Antrieb provides the infrastructure; the LLM provides the intelligence.

Example Workflow

You: provision 3 ubuntu nodes and install nginx on all nodes


LLM: provision(cluster: ["ubuntu24.04 x3"])
→ { session_id: "abc12", nodes: ["node1", "node2", "node3"], provision_time_ms: 720 }

LLM: exec(session_id: "abc12", node: "node1", command: "apt-get update && apt-get install -y nginx")
→ { exit_code: 0, stdout: "..." }

LLM: exec(session_id: "abc12", node: "node1", command: "systemctl start nginx && curl -s localhost")
→ { exit_code: 0, stdout: "<html>Welcome to nginx...</html>" }

LLM: save(type: "image", session_id: "abc12", node: "node1", name: "my-nginx", commands: [...])
→ { ani: "antrieb:my-nginx:v1" }

LLM: delete(type: "cluster", name: "abc12")
→ { success: true }

Available Images

Base Images

ANI	Description
`ubuntu24.04`	Ubuntu 24.04 LTS: apt, bash, Python 3, curl, wget, jq
`almalinux9`	AlmaLinux 9 (RHEL-compatible): dnf, bash, Python 3
`centos-stream10`	CentOS Stream 10: dnf, bash, Python 3
`debian13`	Debian 13: apt, bash, Python 3, curl, wget, jq
`alpine`	Alpine Linux 3.23: apk, minimal, musl libc
`sonic`	SONiC: open-source network OS for data-center switches
`vyos`	VyOS: Linux-based router / firewall with a unified CLI
`openwrt`	OpenWrt 25.12: Linux-based firmware with UCI configuration
`opnsense`	OPNsense: FreeBSD-based firewall / router with web UI

Stack Images

Stack images are custom images created by enriching the base images with additional packages.

ANI	Description
`terraform-aws`	Terraform with real AWS (free-tier, Antrieb's credentials, Vault-brokered STS)
`ansible-controller`	Ansible control node with collections
`podman`	Podman, Buildah, Skopeo (rootless, Docker-API-compatible)
`mongodb`	MongoDB server image for database provisioning and integration tests
`postgresql`	PostgreSQL server image for SQL database tests
`vault`	HashiCorp Vault image for secrets workflows and Terraform-backed credentials
`mariadb`	MariaDB server image for MySQL-compatible database tests
`nginx`	NGINX image for web server and reverse-proxy scenarios
`helm-k3s`	K3s with Helm for lightweight Kubernetes scenarios
`asterisk-pbx`	Asterisk PBX image for SIP and telephony scenarios

Use search to discover all available images with full descriptions, or just specify a distro name and Antrieb picks the right one.

Cluster Networking

Networking is a first-class input to provision, not an afterthought. Every cluster is L2-isolated from every other cluster (even ones you own), with hostname resolution, passwordless SSH between its own nodes, and cluster-scoped env vars auto-injected into every exec call. What varies between runs is how nodes are wired and controlled by the optional networks and nics parameters on provision.

There are two modes: the default network (flat, one-subnet, internet-reachable) and declared networks (first-class multi-subnet, multi-NIC topologies).

Every cluster (both modes)

Hostname resolution: node1, node2, node3 in /etc/hosts on every node.
Passwordless SSH: a per-cluster ed25519 keypair distributed to all nodes. ssh node2 hostname just works.
L2 isolation: nodes across different clusters cannot reach each other, regardless of CIDR overlap or account.
Env vars on every exec: NODE_NAME, NODE_INDEX, NODE_IP, CLUSTER_HOSTS (multiline IP NAME — drop-in for /etc/hosts), CLUSTER_SSH_PUBKEY, CLUSTER_SSH_PRIVKEY.

LLM: exec(node: "node1", command: "ssh node2 hostname")
→ { stdout: "node2" }

Default network (when `networks` / `nics` are omitted)

LLM: provision(cluster: ["ubuntu24.04 x3"])

You get:

One auto-allocated /24 network named default, one NIC per node, DHCP-assigned IPs.
Egress to the internet via a curated port allowlist (HTTP, HTTPS, DNS, SSH, package managers). Arbitrary outbound is blocked.
No router VM, no firewall zones, no topology to configure. Nodes talk to each other and to the outside world.

Use this whenever the network isn't what you're testing — application-level work, distro-specific configuration, Ansible playbooks, CI pipelines, Kubernetes-on-three-ubuntus.

Declared networks (multi-network, multi-NIC)

Pass networks and nics when the topology itself is the thing under test: routers, firewalls, DMZs, isolated back-end subnets, dual-homed nodes, BGP meshes.

LLM: provision(
  cluster: ["vyos", "ubuntu24.04", "ubuntu24.04"],
  networks: [
    { name: "wan", egress: true },
    { name: "lan", cidr: "10.10.1.0/24", egress: false }
  ],
  nics: {
    node1: [{ net: "wan" }, { net: "lan" }],   // vyos: dual-homed router
    node2: [{ net: "lan" }],                    // ubuntu backend
    node3: [{ net: "lan" }]                     // ubuntu backend
  }
)

Contract:

Each declared network is its own L2 segment with its own DHCP server and its own /24. CIDRs are auto-allocated when omitted; only /24 is supported.
egress: true — internet-reachable on the default port allowlist. egress: false — traffic stays inside the network; no default route out.
dhcp: true (default) — the platform assigns IPs. dhcp: false — you assign IPs (use this when a router VM owns addressing or you need precise /31 links).
NIC order matters. Each entry in nics.<node> becomes eth0, eth1, ... in the listed order.
.1 of every DHCP-enabled network is reserved for the platform bridge gateway. Router VMs should claim .254 (or DHCP-client) — assigning .1 to a VM causes ARP flip-flop.

This unlocks serious networking work: VyOS / OPNsense firewalls with WAN/LAN zones, SONiC as an L2 switch or L3 router, OpenWrt with its fw3 zone model, nginx behind a reverse-proxy LAN, BGP peering over /31 transit links, hairpin NAT, DNAT port-forwards. Full walkthroughs for each pattern live in the @antrieb/* runbook library.

Reusable topologies should be captured in runbooks. Networks are declared inline during provision; if the same topology matters across scenarios, save the topology, commands, and verification steps together as a runbook.

Example Network Labs

The default antrieb/* runbook library includes verified scenarios such as:

VyOS: zone-based DMZ firewall, DNAT port forwarding, DHCP/DNS forwarding, DHCP relay, dual-WAN failover, policy-based routing, eBGP, OSPF, VRRP, WireGuard, IPsec, VLAN trunking, QoS shaping.
SONiC: L2 bridging, L3 routing, VLAN segmentation, inter-VLAN routing with policy, BGP with FRR, OSPF, DHCP relay, LAG, SPAN mirroring, VXLAN L2 stretch.
OPNsense: DNAT port forwarding, three-zone DMZ, WireGuard road warrior, HTTPS load balancing with HAProxy.
OpenWrt: home-router NAT, guest-zone isolation, port forwarding, dnsmasq static DHCP and local DNS overrides.
Service topologies: nginx reverse proxy, Ansible-managed nginx backends with HAProxy load balancing and failover drills.

Runbooks are written as executable proofs: topology, commands, expected packet behavior, and verification steps using tools like tcpdump, route tables, conntrack, BGP/OSPF state, DHCP traces, VPN handshakes, and failover tests.

Runbooks

Runbooks are the way to preserve non-trivial multi-node scenarios, capturing even little gotchas that may be too small to blog about. They're markdown documents with structure the LLM can execute:

## Topology required — cluster shape this applies to (images, networks, NICs)
## Variables — placeholders like <LAN_SUBNET> callers substitute at apply time
## Steps — numbered; prose explains why, fenced shell blocks are the actual commands, each annotated with the node it targets
## Verify — how to confirm success

Save one with save(type: "runbook", session_id: "...", name: "...", body: "...markdown..."). The session_id must point to a session whose exec history demonstrates the runbook. Fetch the body with search(type: "runbook", fq_name: "namespace/name").

Runbooks complement images: an image captures installed state (packages, configs baked into a qcow2). A runbook captures a workflow (the ordered, multi-node actions that produce a working system). Prefer runbooks for anything with more than one node, because a single image can't express coordination between VMs.

Everything you save lives in your org's namespace; the antrieb/* namespace is read-only defaults.

Custom Images

Save any configured node as a reusable image:

Provision a base image
Install and configure software via exec
Call save(type: "image", ...) with the list of successful commands
Antrieb generates build-image.sh, startup.sh, and a description
The image is immediately available for future provision calls

Try It Now

No MCP client needed. Use your API key with curl:

# 1. Provision a single Ubuntu node
curl -s -X POST https://antrieb.sh/mcp \
  -H "Authorization: Bearer ant_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"provision","arguments":{"cluster":["ubuntu24.04"]}}}'

→ { "session_id": "abc12...", "nodes": ["node1"], "provision_time_ms": 680 }

# 2. Run a command on the node
curl -s -X POST https://antrieb.sh/mcp \
  -H "Authorization: Bearer ant_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"exec","arguments":{"session_id":"abc12...","node":"node1","command":"cat /etc/os-release | head -3"}}}'

→ { "node": "node1", "exit_code": 0, "stdout": "PRETTY_NAME=\"Ubuntu 24.04 LTS\"\nNAME=\"Ubuntu\"\nVERSION_ID=\"24.04\"" }

# 3. Tear it down
curl -s -X POST https://antrieb.sh/mcp \
  -H "Authorization: Bearer ant_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"delete","arguments":{"type":"cluster","name":"abc12..."}}}'

Replace ant_YOUR_KEY with your API key and abc12... with the session_id from step 1.

FAQ

Why Antrieb?

AI is moving into every layer of operations: cloud, on-prem, legacy systems, security, networking, and edge. Misconfiguring a cloud server is painful. Misconfiguring ten thousand edge devices in the field is catastrophic.

Antrieb is the validation layer between LLMs and Operations. Before AI-generated infrastructure code can touch your environment, it must prove itself in a real, disposable environment that matches the target architecture closely enough to expose operational failure. Same OS, same packages, same behavior. Not a container. Not an approximation. Antrieb is an auditable proving ground for AI-generated operational change.

Why 10-minute clusters?

The 10-minute TTL is not primarily about compute limits. Clusters are fast to recreate. The TTL exists to make validation reproducible and to keep LLMs from accumulating accidental state.

Don't preserve the lab. Preserve the learning.

State drift: the LLM cannot rely on leftover files, partial fixes, or hidden changes from previous attempts. Each solution must work from a clean slate.
Reproducibility: If a solution cannot be reproduced on a fresh cluster, it does not count.
Cognitive drift: long-lived clusters let an LLM overcommit to its first plan. It may keep patching a broken approach because the environment still exists and contains partial progress. A hard TTL creates a forced checkpoint: the LLM must summarize what it learned, discard accidental state, choose a cleaner command sequence, and reproduce the result from scratch.
Abuse prevention: the TTL also prevents long-running misuse such as scanning, bot activity, mining, or hosting.

Why are clusters immutable?

A cluster's architecture is fixed at provision time: images, node count, networks, and NIC assignments. If the LLM needs a different topology, it should provision a new cluster with a better design.

This prevents architectural drift. Without immutability, an LLM can keep adding nodes, networks, routes, and exceptions until the lab no longer matches the system it was supposed to validate. The result may "work," but only because the environment evolved around the mistakes.

Immutability forces the topology to be an explicit hypothesis:

Choose the architecture.
Provision it.
Test the implementation.
If the architecture was wrong, discard it and provision a cleaner one.

That keeps the validation target understandable, repeatable, and worth saving as a runbook.

Is it free?

Yes. Get a free API key at antrieb.sh/dash.

What are the resource specs of each VM?

Most Linux base and stack images get 4 vCPUs, 2 GB of RAM, and 20 GB of disk. Network appliances and smaller edge images can differ; search(type: "image") returns the current specs and full image descriptions.

How many nodes can a cluster have?

Up to 4 nodes per cluster. If your topology requires a dedicated controller node (for example, an Ansible control node managing a fleet), one additional node is allowed, bringing the maximum to 5.

How many clusters can I run at once?

Up to 2 concurrent clusters per account.

Do nodes have internet access?

Yes, by default. Every node on the default network has direct internet access, though for security reasons only certain ports are allowed. Package managers, curl, pip, npm: all work as you'd expect. If you declare explicit networks with egress: false, nodes on those networks are fully isolated from the internet.

How long do clusters last?

Authenticated clusters have a hard TTL of 10 minutes. After that they are fully discarded: compute, networking, everything. TTLs cannot be extended. If you ever get the same IP address as a previous session, the VM is completely fresh.

What happens to a cluster when my chat session ends?

The cluster lives until you delete it or its TTL expires. IP addresses can then be reused for future VMs. Your commands are logged separately and remain accessible in your dashboard.

Are VMs truly ephemeral? Could I ever get a dirty node?

Yes, fully ephemeral. Session VMs are never reused between sessions. Once you're done or the TTL expires, they are destroyed. Every node you provision is completely fresh.

Who can see the commands running on my VMs?

Only you. No one else has access to your command history.

Can my LLM accidentally expose sensitive data in command logs?

Please do not include sensitive data such as secrets or credentials in your commands. Command logs are accessible to you via the dashboard, so treat them accordingly.

Are my VMs isolated from other users?

Yes. Nodes within a cluster can reach each other, but a node cannot reach a node in a different cluster, including other clusters you own. To verify this yourself, tell your LLM: "Provision two single-node clusters, get each node's IP, then SSH into each and try to ping the other." You'll see the ping fail.

Can I SSH into a node myself, not just through the LLM?

No. All interactions go through the LLM. Tell it what you want to do and it will do it for you. Everything is conversational.

What data is logged? Do you store my commands?

Yes. All commands are logged and made available to you in your dashboard at antrieb.sh/dash.

Which clients does this work with?

Claude.ai, ChatGPT Web, Claude Desktop, Cursor, Windsurf, and Claude Code. If your client supports remote MCP servers over HTTP, it should work.

Does it work with models other than Claude?

Yes. Antrieb is just an MCP server; the intelligence comes entirely from your LLM. It works with any model your MCP client supports.

How long does a saved image take to be ready?

We target 2 minutes. The maximum is 5 minutes. Runbooks save instantly — they're metadata, not qcow2 images.

Are custom images and runbooks private?

Yes, private by default. To share within a team, go to your profile at antrieb.sh/dash and set a namespace for your organization. From that point on, everything you save is accessible only to members of your org.

How does Antrieb prevent abuse?

Command logging and AI monitoring. Every command executed on every node is logged. We use AI to aggressively detect and prevent abuse, including crypto mining, DDoS, spam, and unauthorized scanning.
Cluster limits. Maximum 2 concurrent clusters per user. Up to 4 nodes per cluster.
Cluster-to-cluster isolation. Nodes within a cluster can communicate. Nodes across clusters cannot, even within the same account.
Egress port allowlist. Outbound traffic on the default network is restricted to a curated set of ports (HTTP, HTTPS, DNS, SSH, package managers). Arbitrary outbound connections are blocked. Networks declared with egress: false have no outbound path at all.

How is this different from E2B, Morph?

vs E2B: E2B is excellent for code execution sandboxes. It's optimized for running code, not infrastructure. You get a single sandbox, not a cluster, and the environment doesn't match yours at the OS level. If you're testing infra (Ansible playbooks, systemd behavior, SELinux policies, network), you need the real OS and real packages.

vs Morph Cloud: Morph gives your agent a dev environment. Morph's devboxes are great for coding agents building and iterating on software. Antrieb is for infrastructure work, where fidelity to your actual distro matters: Ubuntu, AlmaLinux, Alpine, Arch, SONiC, the same OS you actually run, not a customized devbox.

How is this different from just using a Docker container or GitHub Codespaces?

Docker containers share a host kernel, so you can't reliably test SELinux policies, kernel modules, systemd behavior, or distro-specific package quirks in a container. Antrieb gives your LLM real VMs running the exact same OS you'd run in your environment.

Beyond Linux distros, many network appliances and infrastructure platforms (SONiC, Cisco IOS-XE, Palo Alto, F5, Juniper) only ship as VM images or have limited container availability. When a container version exists, it is almost always a subset of the real thing: missing features, different APIs, incomplete behavior. Tracking what each vendor's containerized version actually supports becomes a problem in itself. With VMs, you run the real appliance image and get full fidelity.

Codespaces is built for humans. You open a browser or editor, click around, and type. Antrieb is built for LLMs to validate their work. Your LLM provisions a cluster, runs commands, reads output, and iterates, all through tool calls in the same conversation. It's the difference between giving your LLM a screenshot to click through versus a direct API.

Tool Reference

`provision`

Spin up a VM cluster. Returns a session handle, node names, TTL, and how long the provision took. Nodes get private IPs, /etc/hosts entries, and a shared ed25519 keypair for passwordless SSH.

Parameter	Type	Required	Description
`cluster`	array	yes	VM topology. Each entry is either a shorthand string (`"ubuntu24.04"` = 1 VM, `"ubuntu24.04 x3"` = 3 VMs) or an object `{image, count?}`.
`networks`	array	no	Per-cluster private networks. Each is `{name, cidr?, egress?, dhcp?}` or a string `@namespace/name` referring to a saved spec. Default: one network named `default` with NAT egress.
`nics`	object	no	Per-node NIC assignments keyed by node name, e.g. `{ "node1": [{"net":"lan"}] }`. Default: every node gets one NIC on the first network.

Returns session_id, nodes, provision_time_ms, ttl_seconds, expires_at.

`exec`

Run a shell command on a specific node. Returns stdout, stderr, and exit code. Every command has access to cluster env vars (NODE_NAME, NODE_INDEX, NODE_IP, CLUSTER_HOSTS, CLUSTER_SSH_PUBKEY, CLUSTER_SSH_PRIVKEY) for coordination across nodes.

Parameter	Type	Required	Description
`session_id`	string	yes	From `provision`
`node`	string	yes	Node name (e.g. `"node1"`)
`command`	string	yes	Literal shell command to execute

`save`

Persist an artifact. Saved artifacts live in your org's namespace and are immediately available to future provision / search calls.

Parameter	Type	Required	Description
`type`	string	no	`"image"` (default) or `"runbook"`
`name`	string	yes	Lowercase-and-hyphens identifier (becomes `antrieb:<name>:v1` for images, `@<namespace>/<name>` for runbooks)
`description`	string	no	Short human description used by search (strongly recommended for runbooks — it's what the LLM sees in browse mode)
`session_id`	string	if `type=image` or `type=runbook`	For images, the session containing the node to snapshot. For runbooks, the session whose exec history demonstrates the scenario.
`node`	string	if `type=image`	Node to save
`commands`	array	if `type=image`	Ordered list of successful commands executed on the node
`body`	string	if `type=runbook`	Markdown document (see Runbooks for structure)

`search`

Browse catalogs, or fetch a specific runbook. Two modes: browse (keywords or no params → list of metadata) and fetch (type=runbook, fq_name=namespace/name → single runbook including full body, required before applying it).

Parameter	Type	Required	Description
`type`	string	no	`"image"` (default), `"cluster"`, or `"runbook"`
`keywords`	string	no	Full-text filter on name, description, and fq_name
`fq_name`	string	no	Fetch a specific runbook by fully-qualified name and return its full markdown body. Ignored for images and clusters.
`limit`	number	no	Max results (default 20, max 100)

`delete`

Destroy a cluster or decommission a saved artifact. antrieb/* defaults are never deletable.

Parameter	Type	Required	Description
`type`	string	yes	`"cluster"`, `"image"`, or `"runbook"`
`name`	string	yes	`session_id` for clusters; short name or `namespace/short` fq_name for everything else

License

Apache 2.0

Featured

CodeRabbit

AI writes the code. CodeRabbit catches the slop.

Try For Free →

Keep your Mac awake

Keep your Mac awake while Claude Code and 40+ AI agents run. Sleeps when they're idle.

One time payment $9 →

Context.dev

Integrate web data into your AI product. One API to scrape website & brand data.

Get API Key Now →

Make your agent a DeFi expert

Agent, run crypto. Access onchain data & trade routes via 1inch.

Install now →

Make money from your Skills

On Capafy, your Skill runs online 24/7 as an agent product, and you get paid every time someone uses it.

Start earning →

AppSignal

Monitor with ease. Code with confidence.

Start Free Trial →

Registryactive

TransportHTTP

UpdatedMar 28, 2026

View on GitHub

Antrieb MCP Server

Instant disposable clusters for LLM-generated networks and infra.

Root access, private networking, passwordless SSH between nodes. Ten minutes per cluster. Clean slate every time. Instant clusters that change the economics of being wrong for LLMs.

Quick Start

Antrieb is a remote MCP server. Nothing to install. Add it to your config and start provisioning.

Claude Web / ChatGPT

With web-based AI clients, add Antrieb as a custom connector.

In claude.ai: go to Settings → Connectors, click Add custom connector. Enter:

Name: Antrieb
URL: https://antrieb.sh/mcp

Click Add, then Connect and sign in. Back in your chat, try something like:

Create a 3-node Kubernetes cluster, with node1 running both as control-plane and worker, and node2/node3 as pure workers. Install nginx with 3 pods and pod anti-affinity.

Walkthrough video: https://youtu.be/8nts8ol-yeA

MCP Clients (Claude Desktop, Claude Code, Codex, custom clients)

Get a free API key at antrieb.sh/dash, then add this to your MCP client config (e.g. .mcp.json for Claude Desktop):

Claude Code

Put this JSON in the current folder in a file called .mcp.json. Claude Code will find it automatically.

{
  "mcpServers": {
    "antrieb": {
      "type": "http",
      "url": "https://antrieb.sh/mcp",
      "headers": {
        "Authorization": "Bearer ant_YOUR_API_KEY"
      }
    }
  }
}

Codex

Put this TOML content in ./.codex/config.toml. Then set the ANTRIEB_TOKEN environment variable with your API key before launching Codex.

 [mcp_servers.antrieb]
  url = "https://antrieb.sh/mcp"
  bearer_token_env_var = "ANTRIEB_TOKEN"

No local install, no dependencies, no Docker.

How It Works

Your LLM controls real VMs through 5 tools:

1. provision  →  Spin up a cluster (sub-second per VM, declare custom networks and NICs)
2. exec       →  Run commands on any node (cluster-scoped env vars auto-injected)
3. save       →  Snapshot a node as an image, or save a runbook
4. search     →  Discover images, runbooks, or active clusters
5. delete     →  Destroy a cluster, or decommission a saved artifact

Example Workflow

You: provision 3 ubuntu nodes and install nginx on all nodes


LLM: provision(cluster: ["ubuntu24.04 x3"])
→ { session_id: "abc12", nodes: ["node1", "node2", "node3"], provision_time_ms: 720 }

LLM: exec(session_id: "abc12", node: "node1", command: "apt-get update && apt-get install -y nginx")
→ { exit_code: 0, stdout: "..." }

LLM: exec(session_id: "abc12", node: "node1", command: "systemctl start nginx && curl -s localhost")
→ { exit_code: 0, stdout: "<html>Welcome to nginx...</html>" }

LLM: save(type: "image", session_id: "abc12", node: "node1", name: "my-nginx", commands: [...])
→ { ani: "antrieb:my-nginx:v1" }

LLM: delete(type: "cluster", name: "abc12")
→ { success: true }

Available Images

Base Images

ANI	Description
`ubuntu24.04`	Ubuntu 24.04 LTS: apt, bash, Python 3, curl, wget, jq
`almalinux9`	AlmaLinux 9 (RHEL-compatible): dnf, bash, Python 3
`centos-stream10`	CentOS Stream 10: dnf, bash, Python 3
`debian13`	Debian 13: apt, bash, Python 3, curl, wget, jq
`alpine`	Alpine Linux 3.23: apk, minimal, musl libc
`sonic`	SONiC: open-source network OS for data-center switches
`vyos`	VyOS: Linux-based router / firewall with a unified CLI
`openwrt`	OpenWrt 25.12: Linux-based firmware with UCI configuration
`opnsense`	OPNsense: FreeBSD-based firewall / router with web UI

Stack Images

Stack images are custom images created by enriching the base images with additional packages.

ANI	Description
`terraform-aws`	Terraform with real AWS (free-tier, Antrieb's credentials, Vault-brokered STS)
`ansible-controller`	Ansible control node with collections
`podman`	Podman, Buildah, Skopeo (rootless, Docker-API-compatible)
`mongodb`	MongoDB server image for database provisioning and integration tests
`postgresql`	PostgreSQL server image for SQL database tests
`vault`	HashiCorp Vault image for secrets workflows and Terraform-backed credentials
`mariadb`	MariaDB server image for MySQL-compatible database tests
`nginx`	NGINX image for web server and reverse-proxy scenarios
`helm-k3s`	K3s with Helm for lightweight Kubernetes scenarios
`asterisk-pbx`	Asterisk PBX image for SIP and telephony scenarios

Use search to discover all available images with full descriptions, or just specify a distro name and Antrieb picks the right one.

Cluster Networking

There are two modes: the default network (flat, one-subnet, internet-reachable) and declared networks (first-class multi-subnet, multi-NIC topologies).

Every cluster (both modes)

Hostname resolution: node1, node2, node3 in /etc/hosts on every node.
Passwordless SSH: a per-cluster ed25519 keypair distributed to all nodes. ssh node2 hostname just works.
L2 isolation: nodes across different clusters cannot reach each other, regardless of CIDR overlap or account.
Env vars on every exec: NODE_NAME, NODE_INDEX, NODE_IP, CLUSTER_HOSTS (multiline IP NAME — drop-in for /etc/hosts), CLUSTER_SSH_PUBKEY, CLUSTER_SSH_PRIVKEY.

LLM: exec(node: "node1", command: "ssh node2 hostname")
→ { stdout: "node2" }

Default network (when `networks` / `nics` are omitted)

LLM: provision(cluster: ["ubuntu24.04 x3"])

You get:

One auto-allocated /24 network named default, one NIC per node, DHCP-assigned IPs.
Egress to the internet via a curated port allowlist (HTTP, HTTPS, DNS, SSH, package managers). Arbitrary outbound is blocked.
No router VM, no firewall zones, no topology to configure. Nodes talk to each other and to the outside world.

Use this whenever the network isn't what you're testing — application-level work, distro-specific configuration, Ansible playbooks, CI pipelines, Kubernetes-on-three-ubuntus.

Declared networks (multi-network, multi-NIC)

Pass networks and nics when the topology itself is the thing under test: routers, firewalls, DMZs, isolated back-end subnets, dual-homed nodes, BGP meshes.

LLM: provision(
  cluster: ["vyos", "ubuntu24.04", "ubuntu24.04"],
  networks: [
    { name: "wan", egress: true },
    { name: "lan", cidr: "10.10.1.0/24", egress: false }
  ],
  nics: {
    node1: [{ net: "wan" }, { net: "lan" }],   // vyos: dual-homed router
    node2: [{ net: "lan" }],                    // ubuntu backend
    node3: [{ net: "lan" }]                     // ubuntu backend
  }
)

Contract:

Each declared network is its own L2 segment with its own DHCP server and its own /24. CIDRs are auto-allocated when omitted; only /24 is supported.
egress: true — internet-reachable on the default port allowlist. egress: false — traffic stays inside the network; no default route out.
dhcp: true (default) — the platform assigns IPs. dhcp: false — you assign IPs (use this when a router VM owns addressing or you need precise /31 links).
NIC order matters. Each entry in nics.<node> becomes eth0, eth1, ... in the listed order.
.1 of every DHCP-enabled network is reserved for the platform bridge gateway. Router VMs should claim .254 (or DHCP-client) — assigning .1 to a VM causes ARP flip-flop.

Example Network Labs

The default antrieb/* runbook library includes verified scenarios such as:

VyOS: zone-based DMZ firewall, DNAT port forwarding, DHCP/DNS forwarding, DHCP relay, dual-WAN failover, policy-based routing, eBGP, OSPF, VRRP, WireGuard, IPsec, VLAN trunking, QoS shaping.
SONiC: L2 bridging, L3 routing, VLAN segmentation, inter-VLAN routing with policy, BGP with FRR, OSPF, DHCP relay, LAG, SPAN mirroring, VXLAN L2 stretch.
OPNsense: DNAT port forwarding, three-zone DMZ, WireGuard road warrior, HTTPS load balancing with HAProxy.
OpenWrt: home-router NAT, guest-zone isolation, port forwarding, dnsmasq static DHCP and local DNS overrides.
Service topologies: nginx reverse proxy, Ansible-managed nginx backends with HAProxy load balancing and failover drills.

Runbooks

Runbooks are the way to preserve non-trivial multi-node scenarios, capturing even little gotchas that may be too small to blog about. They're markdown documents with structure the LLM can execute:

## Topology required — cluster shape this applies to (images, networks, NICs)
## Variables — placeholders like <LAN_SUBNET> callers substitute at apply time
## Steps — numbered; prose explains why, fenced shell blocks are the actual commands, each annotated with the node it targets
## Verify — how to confirm success

Everything you save lives in your org's namespace; the antrieb/* namespace is read-only defaults.

Custom Images

Save any configured node as a reusable image:

Provision a base image
Install and configure software via exec
Call save(type: "image", ...) with the list of successful commands
Antrieb generates build-image.sh, startup.sh, and a description
The image is immediately available for future provision calls

Try It Now

No MCP client needed. Use your API key with curl:

# 1. Provision a single Ubuntu node
curl -s -X POST https://antrieb.sh/mcp \
  -H "Authorization: Bearer ant_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"provision","arguments":{"cluster":["ubuntu24.04"]}}}'

→ { "session_id": "abc12...", "nodes": ["node1"], "provision_time_ms": 680 }

# 2. Run a command on the node
curl -s -X POST https://antrieb.sh/mcp \
  -H "Authorization: Bearer ant_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"exec","arguments":{"session_id":"abc12...","node":"node1","command":"cat /etc/os-release | head -3"}}}'

→ { "node": "node1", "exit_code": 0, "stdout": "PRETTY_NAME=\"Ubuntu 24.04 LTS\"\nNAME=\"Ubuntu\"\nVERSION_ID=\"24.04\"" }

# 3. Tear it down
curl -s -X POST https://antrieb.sh/mcp \
  -H "Authorization: Bearer ant_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"delete","arguments":{"type":"cluster","name":"abc12..."}}}'

Replace ant_YOUR_KEY with your API key and abc12... with the session_id from step 1.

FAQ

Why Antrieb?

Why 10-minute clusters?

The 10-minute TTL is not primarily about compute limits. Clusters are fast to recreate. The TTL exists to make validation reproducible and to keep LLMs from accumulating accidental state.

Don't preserve the lab. Preserve the learning.

State drift: the LLM cannot rely on leftover files, partial fixes, or hidden changes from previous attempts. Each solution must work from a clean slate.
Reproducibility: If a solution cannot be reproduced on a fresh cluster, it does not count.
Cognitive drift: long-lived clusters let an LLM overcommit to its first plan. It may keep patching a broken approach because the environment still exists and contains partial progress. A hard TTL creates a forced checkpoint: the LLM must summarize what it learned, discard accidental state, choose a cleaner command sequence, and reproduce the result from scratch.
Abuse prevention: the TTL also prevents long-running misuse such as scanning, bot activity, mining, or hosting.

Why are clusters immutable?

A cluster's architecture is fixed at provision time: images, node count, networks, and NIC assignments. If the LLM needs a different topology, it should provision a new cluster with a better design.

Immutability forces the topology to be an explicit hypothesis:

Choose the architecture.
Provision it.
Test the implementation.
If the architecture was wrong, discard it and provision a cleaner one.

That keeps the validation target understandable, repeatable, and worth saving as a runbook.

Is it free?

Yes. Get a free API key at antrieb.sh/dash.

What are the resource specs of each VM?

How many nodes can a cluster have?

Up to 4 nodes per cluster. If your topology requires a dedicated controller node (for example, an Ansible control node managing a fleet), one additional node is allowed, bringing the maximum to 5.

How many clusters can I run at once?

Up to 2 concurrent clusters per account.

Do nodes have internet access?

How long do clusters last?

What happens to a cluster when my chat session ends?

The cluster lives until you delete it or its TTL expires. IP addresses can then be reused for future VMs. Your commands are logged separately and remain accessible in your dashboard.

Are VMs truly ephemeral? Could I ever get a dirty node?

Yes, fully ephemeral. Session VMs are never reused between sessions. Once you're done or the TTL expires, they are destroyed. Every node you provision is completely fresh.

Who can see the commands running on my VMs?

Only you. No one else has access to your command history.

Can my LLM accidentally expose sensitive data in command logs?

Please do not include sensitive data such as secrets or credentials in your commands. Command logs are accessible to you via the dashboard, so treat them accordingly.

Are my VMs isolated from other users?

Can I SSH into a node myself, not just through the LLM?

No. All interactions go through the LLM. Tell it what you want to do and it will do it for you. Everything is conversational.

What data is logged? Do you store my commands?

Yes. All commands are logged and made available to you in your dashboard at antrieb.sh/dash.

Which clients does this work with?

Claude.ai, ChatGPT Web, Claude Desktop, Cursor, Windsurf, and Claude Code. If your client supports remote MCP servers over HTTP, it should work.

Does it work with models other than Claude?

Yes. Antrieb is just an MCP server; the intelligence comes entirely from your LLM. It works with any model your MCP client supports.

How long does a saved image take to be ready?

We target 2 minutes. The maximum is 5 minutes. Runbooks save instantly — they're metadata, not qcow2 images.

Are custom images and runbooks private?

How does Antrieb prevent abuse?

Command logging and AI monitoring. Every command executed on every node is logged. We use AI to aggressively detect and prevent abuse, including crypto mining, DDoS, spam, and unauthorized scanning.
Cluster limits. Maximum 2 concurrent clusters per user. Up to 4 nodes per cluster.
Cluster-to-cluster isolation. Nodes within a cluster can communicate. Nodes across clusters cannot, even within the same account.
Egress port allowlist. Outbound traffic on the default network is restricted to a curated set of ports (HTTP, HTTPS, DNS, SSH, package managers). Arbitrary outbound connections are blocked. Networks declared with egress: false have no outbound path at all.

How is this different from E2B, Morph?

How is this different from just using a Docker container or GitHub Codespaces?

Tool Reference

`provision`

Spin up a VM cluster. Returns a session handle, node names, TTL, and how long the provision took. Nodes get private IPs, /etc/hosts entries, and a shared ed25519 keypair for passwordless SSH.

Parameter	Type	Required	Description
`cluster`	array	yes	VM topology. Each entry is either a shorthand string (`"ubuntu24.04"` = 1 VM, `"ubuntu24.04 x3"` = 3 VMs) or an object `{image, count?}`.
`networks`	array	no	Per-cluster private networks. Each is `{name, cidr?, egress?, dhcp?}` or a string `@namespace/name` referring to a saved spec. Default: one network named `default` with NAT egress.
`nics`	object	no	Per-node NIC assignments keyed by node name, e.g. `{ "node1": [{"net":"lan"}] }`. Default: every node gets one NIC on the first network.

Returns session_id, nodes, provision_time_ms, ttl_seconds, expires_at.

`exec`

Parameter	Type	Required	Description
`session_id`	string	yes	From `provision`
`node`	string	yes	Node name (e.g. `"node1"`)
`command`	string	yes	Literal shell command to execute

`save`

Persist an artifact. Saved artifacts live in your org's namespace and are immediately available to future provision / search calls.

Parameter	Type	Required	Description
`type`	string	no	`"image"` (default) or `"runbook"`
`name`	string	yes	Lowercase-and-hyphens identifier (becomes `antrieb:<name>:v1` for images, `@<namespace>/<name>` for runbooks)
`description`	string	no	Short human description used by search (strongly recommended for runbooks — it's what the LLM sees in browse mode)
`session_id`	string	if `type=image` or `type=runbook`	For images, the session containing the node to snapshot. For runbooks, the session whose exec history demonstrates the scenario.
`node`	string	if `type=image`	Node to save
`commands`	array	if `type=image`	Ordered list of successful commands executed on the node
`body`	string	if `type=runbook`	Markdown document (see Runbooks for structure)

`search`

Parameter	Type	Required	Description
`type`	string	no	`"image"` (default), `"cluster"`, or `"runbook"`
`keywords`	string	no	Full-text filter on name, description, and fq_name
`fq_name`	string	no	Fetch a specific runbook by fully-qualified name and return its full markdown body. Ignored for images and clusters.
`limit`	number	no	Max results (default 20, max 100)

`delete`

Destroy a cluster or decommission a saved artifact. antrieb/* defaults are never deletable.

Parameter	Type	Required	Description
`type`	string	yes	`"cluster"`, `"image"`, or `"runbook"`
`name`	string	yes	`session_id` for clusters; short name or `namespace/short` fq_name for everything else

License

Apache 2.0

Antrieb

Antrieb MCP Server

Quick Start

Claude Web / ChatGPT

MCP Clients (Claude Desktop, Claude Code, Codex, custom clients)

Claude Code

Codex

How It Works

Example Workflow

Available Images

Base Images

Stack Images

Cluster Networking

Every cluster (both modes)

Default network (when networks / nics are omitted)

Declared networks (multi-network, multi-NIC)

Example Network Labs

Runbooks

Custom Images

Try It Now

FAQ

Tool Reference

provision

exec

save

search

delete

License

Antrieb

Antrieb MCP Server

Quick Start

Claude Web / ChatGPT

MCP Clients (Claude Desktop, Claude Code, Codex, custom clients)

Claude Code

Codex

How It Works

Example Workflow

Available Images

Base Images

Stack Images

Cluster Networking

Every cluster (both modes)

Default network (when networks / nics are omitted)

Declared networks (multi-network, multi-NIC)

Example Network Labs

Runbooks

Custom Images

Try It Now

FAQ

Tool Reference

provision

exec

save

search

delete

License

Default network (when `networks` / `nics` are omitted)

`provision`

`exec`

`save`

`search`

`delete`

Default network (when `networks` / `nics` are omitted)

`provision`

`exec`

`save`

`search`

`delete`