generated: '2026-09-05'
method: searched
status: candidate
source: >-
  searched the Cat Digital Marketplace, the CaterpillarInc GitHub organization,
  npm (@modelcontextprotocol and vendor scopes) and public MCP registries, 2026-09-05
note: >-
  Caterpillar ships NO Model Context Protocol server — no hosted endpoint, no
  stdio package, no reference in Cat Digital documentation or the CaterpillarInc
  GitHub organization (which has zero public repositories). The tools below are a
  CANDIDATE surface derived from the endpoints Caterpillar documents publicly; they
  do not exist as MCP tools and no endpoint is asserted. Input schemas are not
  recorded because Caterpillar publishes no OpenAPI to inherit parameters from —
  writing them would be invention. This artifact is pointed at from apis.yml with
  type X-MCPServerCandidate, not MCPServer, precisely so it is not credited as a
  shipped agent surface.
deployment:
  mode: none
  endpoint: null
  install: null
  package: null
  auth: oauth
  verified: searched
server:
  name: null
  transport: null
  url: null
tools:
  - name: get_fleet_snapshot
    description: Retrieve a paginated ISO 15143-3 (AEMP 2.0) fleet snapshot.
    source_endpoint: GET https://services.cat.com/telematics/iso15143/fleet/{pageNumber}
    input_schema: null
    confidence: medium
  - name: get_equipment_snapshot
    description: Retrieve an ISO 15143-3 (AEMP 2.0) snapshot for one machine.
    source_endpoint: GET https://services.cat.com/telematics/iso15143/equipment/{equipmentId}
    input_schema: null
    confidence: medium
  - name: get_asset_summary
    description: Retrieve VisionLink asset summary data.
    source_endpoint: GET https://services.cat.com/catDigital/assetSummary/v1/assets
    input_schema: null
    confidence: medium
  - name: get_fault_history
    description: Retrieve VisionLink fault history for machine health analysis (source, type, severity).
    source_endpoint: GET https://services.cat.com/catDigital/faultsHistory/v1/faults
    input_schema: null
    confidence: medium
agent_readiness_gaps:
  - No MCP server of any deployment mode.
  - No public OpenAPI, so no tool can inherit a real inputSchema.
  - No /.well-known/oauth-protected-resource, so a token holder cannot discover the resource server.
  - No rate-limit response headers, so an agent cannot read its remaining budget at runtime.
  - No sandbox and no dry-run mode, so nothing can be rehearsed before it is called.