FactVerse MCP Integration

FactVerse uses Model Context Protocol to connect AI agents with governed industrial tools, digital twin context, simulation services, operational data, and approved execution paths.

Integration model

An MCP-compatible client sends requests to a FactVerse MCP endpoint with an API key. FactVerse resolves the tenant and allowed scopes from the key, then exposes only the tools that the key is allowed to use.

Prerequisites

Before connecting a client, prepare:

• endpoint family such as /mcp/base/, /mcp/pdm/, or another enabled module endpoint;
• API key owner, tenant, scopes, rotation rule, and revocation path;
• customer workflow boundary and reviewer;
• source data readiness notes when the workflow depends on DFS, documents, work records, scenes, or module records;
• run-record location for evidence, assumptions, approvals, and feedback.

Authentication

Customer-facing MCP requests use an API key in the request header:

X-API-Key: your-api-key

The key identifies the tenant and the scopes assigned to that customer or integration. Tenant context resolves from the issued key.

Example client configuration

{
  "mcpServers": {
    "factverse": {
      "url": "https://your-factverse-host/mcp/base/",
      "headers": {
        "X-API-Key": "your-api-key"
      }
    }
  }
}

Working with tools

Use the MCP client to list available tools at runtime. Tool availability can change as modules, customer scopes, and customer environment configuration evolve.

Use the Tool Reference for stable tool names, scopes, descriptions, and parameters. Use runtime tool discovery to confirm what a specific customer environment and API key can call.

Common setup path

1. Choose the customer workflow and operating owner.
2. Select the endpoint family that matches the workflow.
3. Request the minimum API key scopes.
4. Configure the MCP client with the endpoint URL and X-API-Key.
5. Initialize the client and list visible tools.
6. Run a read-only tool call with a small test request.
7. Confirm source timestamps, identifiers, and assumptions appear in the output.
8. Record the test result before adding compute or draft-action workflows.

Validation checklist

• Client can initialize the MCP session.
• Tool discovery returns the expected tools for the key.
• Refused calls fail closed when scope is missing.
• Source timestamps and identifiers are visible in the output.
• Compute and write actions follow the configured review path.
• Errors are recorded with endpoint, tool, scope, and request context.

Failure handling

Problem	Check
Client cannot connect	Endpoint URL, HTTPS route, trailing slash, network allowlist, and server availability.
Authentication fails	API key value, tenant assignment, key revocation state, and header name.
Tool list is empty	Endpoint family, scopes, module enablement, and runtime discovery.
Tool call is refused	Required scope, package gate, and tenant boundary.
Output lacks evidence	Source data readiness, DFS mappings, document access, scene version, or module records.

Read next

Page	Use
Integration Guide	Configure Claude, Cursor, Python, or another MCP-compatible client.
Capability Guide	Understand endpoint families, capability categories, and workflow selection.
Scope Matrix	Plan endpoint, scope, reviewer, and access package decisions.
Errors and Audit	Diagnose access, scope, source-data, compute, and write-action failures.
Tool Reference	Check tool names, scopes, descriptions, and parameters.
Scope Reference	Check endpoint and scope details.

Physical AI endpoint planning

Physical AI integrations use the base platform and module-specific MCP endpoints described in the scope reference.