What it has to do
Five jobs an MCP control plane must do on every call
1. Know what exists. An inventory of every MCP server and every tool it exposes, internal and external, with drift detection when a server changes its tools after you approved them.
2. Know who is calling, and for whom. The agent's identity and, when it acts for a person, that person's identity too, so both are checked.
3. Decide per tool. Permissions at the level of a single tool, not a whole server, so an agent can read a ledger without being able to post to it.
4. Apply your policy. Limits, approvals and conditions on the call's arguments, from your own rules, enforced before the tool runs.
5. Record and price it. A tamper-evident audit trail of every decision, and the cost of every call, attributed to a person and a department.
How Magertron does it
Governance at the data plane, inside your cluster
Every MCP call passes through Magertron's gateway in your cluster. Identity, per-tool permissions, your policy and spend limits are checked there, before the tool runs. Calls an approver must see are held until a named person decides. See how the enforcement layers fit together and how data access is governed.
Questions
Frequently asked
What is an MCP control plane?
An MCP control plane is the layer that decides which AI agents may call which Model Context Protocol servers and tools, under whose identity, within what limits, and records every decision. The MCP servers do the work; the control plane governs who can ask them to.
How is an MCP control plane different from an MCP gateway?
A gateway routes and authenticates traffic. A control plane also keeps the inventory of servers and tools, holds the per-tool permissions, applies your policy, meters and prices each call, and keeps the audit trail. Magertron's AI Orchestrator includes the gateway and the control plane.
Where does Magertron's MCP control plane run?
In your own Kubernetes cluster, on any cloud, on-premises or air-gapped. Tool traffic and credentials stay inside your network.
Can I use my own policy engine?
Yes. On Enterprise, Magertron can ask your own policy decision point, such as Open Policy Agent, on every call, with the caller, the person it acts for, the tool and its arguments.
Does it work with Claude Code, Cursor and other agent tools?
Yes. Magertron sits underneath whatever agent harness you use. Agents call MCP servers through it; nothing in the harness changes.