naftiko: 1.0.0-alpha2
info:
  label: Viam ML Model Service — Inference
  description: Run device-side inference against deployed ML models (TFLite, ONNX, Triton).
  tags: [Viam, MachineLearning, Inference, Edge]
  created: '2026-05-25'
  modified: '2026-05-25'
binds:
- namespace: env
  keys:
    VIAM_API_KEY: VIAM_API_KEY
    VIAM_MACHINE_ADDRESS: VIAM_MACHINE_ADDRESS
capability:
  consumes:
  - type: http
    namespace: ml-model-infer
    baseUri: '{{env.VIAM_MACHINE_ADDRESS}}'
    description: Viam MLModelService on a viam-server endpoint.
    resources:
    - name: infer
      path: /viam.service.mlmodel.v1.MLModelService/Infer
      operations:
      - { name: infer, method: POST, description: Run inference against a deployed model., outputRawFormat: json, inputParameters: [{ name: body, in: body, type: object, required: true }] }
    - name: metadata
      path: /viam.service.mlmodel.v1.MLModelService/Metadata
      operations:
      - { name: metadata, method: POST, description: Retrieve the model's tensor schema., outputRawFormat: json, inputParameters: [{ name: body, in: body, type: object, required: true }] }
    authentication:
      type: apikey
      key: key
      value: '{{env.VIAM_API_KEY}}'
      placement: header
  exposes:
  - type: mcp
    namespace: ml-model-infer-mcp
    port: 9090
    transport: http
    description: MCP adapter for device-side ML inference.
    tools:
    - { name: viam-mlmodel-infer, description: Device-side inference., hints: { readOnly: false, destructive: false, idempotent: true }, call: ml-model-infer.infer, with: { body: tools.body } }
    - { name: viam-mlmodel-metadata, description: Get tensor schema., hints: { readOnly: true, destructive: false, idempotent: true }, call: ml-model-infer.metadata, with: { body: tools.body } }