Skip to main content

MCP Capability Guide

FactVerse MCP exposes governed industrial capabilities to MCP-compatible clients. Use this guide to understand how endpoints, scopes, source systems, and generated tool reference pages fit together before selecting tools for a customer workflow.

How to read the MCP model

LayerRole
EndpointA governed MCP slice such as /mcp/base/, /mcp/pdm/, or another enabled module endpoint.
API keyIdentifies the tenant and scopes allowed for that client. The client sends it as X-API-Key.
ScopeControls which tools can be listed and called.
ToolA runtime-discovered function that reads, computes, summarizes, or drafts an approved action.
EvidenceSource references, timestamps, asset IDs, scene versions, and review records returned or recorded by the workflow.

The generated Tool Reference lists the current public catalog for this documentation build. Runtime tool discovery remains the source for what a specific key can call.

Endpoint families

Endpoint familyUse forTypical workflow
/mcp/base/Shared asset reads, document search, data quality checks, simulation, forecasting, optimization, and controlled action drafts.Facility operations, Physical AI preparation, work-order drafting, cross-domain analysis.
/mcp/pdm/Equipment health, anomaly context, predictive maintenance summaries, and component intelligence.Predictive maintenance review and maintenance planning.
/mcp/trafficops/People, vehicle, checkpoint, lane, roster, and surge analysis when the module is enabled.Border operations and traffic flow analysis.
/mcp/telcoops/Network health, incident explanation, and capacity prediction when the module is enabled.Network operations review.
/mcp/semiops/Cleanroom, particle, pressure, utility, SMT, and facility environment signals when the module is enabled.Facility and cleanroom operating review.
/mcp/aviation/Reliability analysis for aviation datasets when the module is enabled.Reliability review and component analysis.

Use the smallest endpoint set that supports the workflow. A client can configure multiple endpoints when a reviewed workflow needs cross-domain evidence.

Capability categories

CategoryWhat it helps withCommon evidence
Asset and knowledge readsUnderstand equipment, documents, procedures, status, and operational context.Asset IDs, document references, timestamps, status records.
Data quality and connectorsCheck whether source data is fresh, mapped, and usable.Connector status, quality flags, missing fields, source timestamps.
Predictive maintenanceReview equipment condition, anomaly windows, and maintenance history.Health output, signal windows, work records, engineer feedback.
Simulation and what-if analysisCompare scenarios, test assumptions, and prepare Physical AI or facility planning work.Scenario ID, input assumptions, backend, output metrics, validation notes.
Optimization and forecastingExplore policy, layout, allocation, demand, or capacity options.Objective, constraints, assumptions, result alternatives.
Spatial analysisAnalyze zones, layouts, paths, sensor placement, and scene-related anomalies.Scene ID, zone, coordinate context, sensor references.
Controlled actionsDraft work orders or operational records after review.Approval state, reviewer, draft record, final action ID.

Workflow selection

Customer taskStart withAdd when needed
Facility status review/mcp/base/ with read scopeCompute scope for approved summaries or analysis.
Predictive maintenance review/mcp/pdm/ with read scope/mcp/base/ for asset documents and reviewed work-order drafts.
Physical AI scene preparation/mcp/base/ with read and compute scopesModule endpoints for operating signals that constrain the scene.
Cleanroom or utility operating review/mcp/semiops/ when enabled/mcp/base/ for documents, work records, and data quality context.
Traffic or checkpoint review/mcp/trafficops/ when enabled/mcp/base/ for shared asset or document context.

Using the generated reference

  1. Choose the workflow and endpoint.
  2. Check Scope Matrix for the minimum scope set.
  3. Use runtime tool discovery from the client.
  4. Cross-check tool names and parameters in the Tool Reference.
  5. Record evidence and review state with the Workflow Run Record.

Good integration behavior

  • List tools at runtime before calling a tool.
  • Keep generated recommendations separate from confirmed facts.
  • Return source timestamps and asset or scene identifiers in user-facing output.
  • Treat compute results as reviewed engineering outputs with visible assumptions.
  • Keep write actions as drafts until a person approves them.
  • Capture reviewer corrections so the workflow improves over time.