MCP server

Planned Limited beta, coming soon

There is no MCP endpoint today: this page describes a design that is planned, not a service you can call.

Back to Developers

The tool surface below is a proposal. Names, inputs and return shapes will change before anything ships.

The constraints further down are product decisions, and they are the part that is meant to hold.

  • Planned 5 items Designed, not started.

How a write would be gated

The tool names below will change. This will not. An agent proposes a change, the twin is used to test it before production sees it, and a human approves it by default. Every stage here is planned, so read it as the shape of the argument rather than a path you can take today.

A gated agent write

Five stages, all planned. Approval is the default, not something you switch on.

  1. Planned

    An agent calls a write tool. The call is a proposal, not an instruction.

  2. Planned

    The call carries a tenant-bound credential. The tenant comes from the credential, never from the arguments, so an agent cannot reach another tenant by asking.

  3. Planned

    The change is tested in memory against the twin first, so its blast radius is known before anything touches production.

  4. Planned

    Approval is the default. Auto-approval is opt-in, per change type, and never the starting state.

  5. Planned

    The write records which agent made it and under which credential. Reads stay free; writes stay auditable.

Proposed tools

Tool names, inputs and return shapes are proposals and will change.

lookup_prefix

Read Planned

Find a prefix and report what it is used for.

Inputs

  • prefix CIDR string optional

    The prefix to look up, such as an exact network.

  • vrf UUID optional

    Narrow the lookup to one VRF.

Returns

  • The matching prefix record.
  • Its site and VRF.
  • Child and parent prefixes.

Constraints on this tool

  • Read-only.
  • Runs under the caller's tenant-bound credential.

allocate_subnet

Write Planned

Propose the next free subnet inside a tenant's address space.

Inputs

  • site UUID required

    The site the subnet is for.

  • vrf UUID required

    The VRF whose address space is searched.

  • size integer required

    Requested prefix length.

Returns

  • A change proposal naming the subnet that would be allocated.
  • The approval state of that proposal.

Constraints on this tool

  • Write tool: the agent proposes, a human approves by default.
  • Auto-approval is opt-in per change type.
  • The record stores which agent and which credential made the request.

inspect_lease

Read Planned

Show the lease history for an address or a MAC address.

Inputs

  • address_or_mac string required

    An IP address or a MAC address in any common format.

Returns

  • Current and past leases for the host.
  • Event timestamps from the ingestion path.

Constraints on this tool

  • Read-only.
  • Runs under the caller's tenant-bound credential.

query_dependencies

Read Planned

Ask the twin what depends on an address.

Inputs

  • address string required

    The address or prefix to trace.

  • depth integer optional

    How far to follow dependencies.

Returns

  • Objects that reference the address.
  • The path from each object to the address.

Constraints on this tool

  • Read-only.
  • Answers come from the twin, not from probing the live network.

what_if

Simulate Planned

Check a proposed change against the twin before anything touches production.

Inputs

  • change object required

    A proposed change, such as a change proposal from a write tool.

Returns

  • Conflicts the change would cause.
  • Objects the change would affect.
  • A pass or fail result.

Constraints on this tool

  • Runs against the twin, in memory.
  • Never writes to production state.

Constraints that are meant to hold

These are product decisions, and they are the part of the design that is meant to hold.

  • Every call runs under a tenant-bound credential

    An MCP call carries the same kind of credential as an API call, and the tenant comes from the credential. The isolation is the same as the API's.

    Why An agent is a caller like any other. Giving it a wider path than a human script would make the agent the weakest point in the tenant boundary.

  • Write tools are gated

    An agent proposes a change and a human approves it by default. Auto-approval is opt-in, per change type.

    Why Network changes are expensive to undo. Starting from human approval means trust is extended one change type at a time, by the people who own the outcome.

  • Reads are free, writes are auditable

    Reads carry no approval step. Every write records which agent and which credential made it.

    Why When something changes, the first question is who did it. A record that names the agent and the credential answers it without guesswork.

  • What-if runs on the twin before production

    A what-if check runs against the twin, in memory, before anything touches production.

    Why Finding a conflict in a model costs nothing. Finding it on the live network costs an outage.

Get access when it ships

Join the beta and you will be told when an MCP endpoint is available.

Join the beta