Skip to main content

Access and Scope Planning

Use this guide before connecting a client or expanding an existing workflow. Access planning should define which endpoint the client uses, which scopes the key holds, who can approve compute or write actions, and how access is reviewed over time.

Planning inputs

InputWhy it matters
Workflow boundaryDefines the tenant, site, asset group, equipment set, scene, or data domain the client can work with.
Endpoint selectionKeeps each client connected to the right governed MCP slice for the task.
Scope setControls which tools are visible and which actions are refused.
Review ownerIdentifies who accepts outputs, approves actions, and reviews exceptions.
Key ownerIdentifies who requests, stores, rotates, and revokes the API key.
Evidence expectationDefines what source references, timestamps, and review notes must appear in the output.

Endpoint selection

Start with the workflow, then choose the smallest endpoint set that supports it.

WorkflowPrimary endpointAdd only when needed
Facility operations/mcp/base/Module endpoints for customer-specific operating signals.
Predictive maintenance/mcp/pdm//mcp/base/ for asset records, documents, and reviewed work-order drafts.
Physical AI/mcp/base/Module endpoints for operational signals that should constrain a scene or simulation task.

Use runtime tool discovery to confirm the final visible tool set for the key. Do not rely only on a static list when building the client workflow.

Scope planning

Apply least privilege first. Add scopes only when a reviewed workflow needs them.

Scope typeTypical usePlanning note
Read scopesAsset context, documents, operating records, health summaries, scene metadata, and source references.First pilots should usually start here.
Compute scopesSimulation preparation, forecasting, optimization, comparison, or summarization jobs.Require an engineering owner for assumptions and output review.
Write scopesDrafting work orders, inspection tasks, review notes, or scenario records.Use only after the approval path and audit record are defined.

When a workflow uses multiple endpoints, document why each endpoint is needed. If one endpoint can satisfy the task, keep the configuration simple.

API key lifecycle

Each key should have an owner, purpose, scope list, and review date.

StageRequired decision
RequestWorkflow name, endpoint, scopes, owner, and expected client.
IssueKey shown once to the owner and configured in the approved client.
UseClient sends X-API-Key on each request and lists tools at runtime.
ReviewOwner checks whether scopes still match the workflow boundary.
RotateReplace the key when ownership, client runtime, or access policy changes.
RevokeRemove the key when the workflow is retired or the client is no longer approved.

Approval boundaries

Separate analysis from action.

Action typeDefault behavior
Read-only summaryAllowed when the key has the matching read scope and the output cites source records.
Compute jobAllowed only when the key has compute scope and the workflow records assumptions, input boundary, and review owner.
Work-order or inspection draftAllowed only when write scope exists and the workflow keeps the record in draft until a person approves it.
Operational changeKeep outside automated execution unless the customer has defined a separate controlled change process.

Access review checklist

  • Endpoint and scopes match the workflow boundary.
  • Runtime tool discovery shows only the expected tools.
  • The client does not pass tenant identifiers as a substitute for key-based authorization.
  • Compute and write actions have named reviewers.
  • Source references and timestamps are visible in accepted outputs.
  • Key owner and review date are documented.
  • Rotation and revocation steps are known to the operating team.