Now available — v4.2.35
NEW: OPA Support
The AI spend plane
for the token economy.
Your agents call tools. Those tools touch data. Magertron governs both — and knows which tool touched which table, for whom, and what it cost. One gateway for model inference and tool calls across every vendor.
Free up to 20 servers — no signup, no credit card.
✦ AI Orchestrator Assistant Advisory · read-only · grounded in live platform state
Trusted by
Install in 5 minutes
# Bare Ubuntu box — installs k3s, helm and Magertron $ curl -fsSL https://magertron.com/bootstrap.sh | sudo bash -s -- \ --api-public-url https://mcp.example.com # Already have a Kubernetes cluster $ git clone https://github.com/magertron/orchestrator.git $ cd orchestrator $ ./install.sh # Free tier — no license needed $ ./install.sh --license /path/to/license.json # With a license
Enterprise Security
Rogue MCP servers are already inside your network
Developers are spinning up ungoverned MCP servers on laptops, in personal cloud accounts, and inside your perimeter — with no audit trail and no controls. AI agents talking to these servers can execute actions at scale.
Now you can find them. Magertron reads MCP client configs on every endpoint you manage, shows you every server in use — approved or not — and points them at the gateway with a single policy change. No resident agent on the laptop — a script and a timer, delivered by the MDM you already run.
THREAT VECTOR 01
External AI agents
AI agents outside your perimeter reaching ungoverned internal MCP servers. Perimeter firewalls catch these — but only if the agent actually crosses the perimeter, and only if the MCP server is registered as a public service.
Governed Surfaces
Every model, API, stream and tool call. One policy plane.
Swipe to see the full diagram →
Enforcement lives in the Envoy ext_authz path.
Every call crosses the same boundary — whether it lands on
an MCP tool, a REST endpoint or an inference provider.
Server Management
Full lifecycle control for every server you run
Deploy MCP servers and LLM pods from any container image, monitor real-time CPU and memory metrics, and scale with a slider. Every action is audited.
- Live metrics charts polling every 5 seconds
- Real pod logs from Kubernetes API
- One-click rollback from deployment history
- Rolling restart with zero downtime
Governance
Enforce policy before anything reaches production
Namespace-scoped governance policies evaluate every deployment at deploy time. Error-severity rules block non-compliant servers. Warnings flag for review.
- Different policies for prod vs dev environments
- Resource limits, transport rules, naming standards
- Dry-run evaluator for pre-flight checks
- Export/import policies across clusters
Capabilities
The full server lifecycle —
in your Kubernetes cluster
Deploy, registry, gateway, governance, observability. One Helm chart. Most platforms cover one or two of these. Magertron covers all five.
⬡
Server Lifecycle
Deploy, scale, restart, and roll back MCP servers and LLM pods from a single console.
▦
Server & Model Registry
Central catalog of every MCP server and LLM provider in your organization. Most platforms don't have a registry — Magertron does, in your cluster.
◈
Governance Engine
Policies for resource limits, transport, naming, and security. Namespace-scoped for prod vs dev.
⛊
Multi-Tenant RBAC
Role-based access with namespace isolation. Admins, deployers, operators, viewers.
⚿
Enterprise Identity
OIDC and SAML SSO. SCIM 2.0 provisioning. Just-in-time accounts. Your IdP is the source of truth.
◷
Audit & Compliance
OCSF-aligned audit events. SIEM-ready, validated with Splunk Cloud. Built for regulated environments.
Developer Experience
A CLI that gets out
of your way
Single binary, zero dependencies. Deploy, scale, and evaluate governance from your terminal. Ships for macOS and Linux.
$ mcpctl login https://mcp.acme.internal admin
✓ Logged in as admin (system:platform-admin)
$ mcpctl deploy code-assistant mcp-prod ghcr.io/acme/code-assistant --tag v2.1 --team platform
Deploying code-assistant to mcp-prod...
✓ Server deployed (state: Running)
$ mcpctl governance evaluate spec.json --namespace mcp-prod
✓ ALLOWED
⚠ Mutating tools detected — review required before production use
$ mcpctl servers
NAME NAMESPACE STATE REPLICAS IMAGE
code-assistant mcp-prod Running 2/2 ghcr.io/acme/code-assistant:v2.1
search-tool mcp-prod Running 3/3 ghcr.io/acme/search:latest
doc-retriever mcp-staging Running 1/1 ghcr.io/acme/docs:v1.4
mcpctl is a signed Go binary — install it below.
There are unrelated mcpctl packages on npm and PyPI; neither is ours.
Homebrew macOS · Linux ▸
Cleanest install if you have brew — and the only one that upgrades itself:
$ brew trust magertron/tap
$ brew install magertron/tap/mcpctl
Homebrew 6 asks you to trust any tap outside its core before it will
load a formula, because a formula is code that runs on your machine.
One command, once. Afterwards brew upgrade keeps
mcpctl current.
curl one-liner macOS · Linux ▸
No package manager required:
$ curl -fsSL https://magertron.com/install-mcpctl.sh | sh
Debian / Ubuntu apt · amd64 · arm64 ▸
One-time setup
$ curl -fsSL https://magertron.com/apt/magertron-archive-keyring.gpg | sudo tee /etc/apt/trusted.gpg.d/magertron-archive-keyring.gpg > /dev/null
$ echo "deb [signed-by=/etc/apt/trusted.gpg.d/magertron-archive-keyring.gpg] https://magertron.com/apt stable main" | sudo tee /etc/apt/sources.list.d/magertron.list
$ sudo apt update
Install
$ sudo apt install mcpctl
RHEL / Fedora / Rocky dnf · x86_64 ▸
One-time setup
$ sudo curl -fsSL https://magertron.com/yum/magertron.repo -o /etc/yum.repos.d/magertron.repo
Install
$ sudo dnf install mcpctl
Or download a binary directly
macOS arm64 · macOS amd64 · Linux amd64 · Linux arm64
Apache 2.0 licensed. Source at
magertron/orchestrator/mcpctl.
Free to download; connecting it to an orchestrator requires a Pro or Enterprise
license — mcpctl login is refused on the Free tier, and commands that
reach higher-tier features return a clear license error naming the
tier they need.
Get a live instance
A private instance, today. Not a slide deck and not a screen share — a URL and a login to a Magertron of your own, with servers, traffic and audit history already running in it. Bring your own model key and the assistant works too.