naftiko: 1.0.0-alpha2
info:
  label: Triton Inference Server NVIDIA Triton Inference Server HTTP/REST API — Inference
  description: 'Triton Inference Server NVIDIA Triton Inference Server HTTP/REST API — Inference. 2 operations. Lead operation:
    Triton Inference Server Run Inference on a Model. Self-contained Naftiko capability covering one Triton business surface.'
  tags:
  - Triton
  - Inference
  created: '2026-05-19'
  modified: '2026-05-19'
binds:
- namespace: env
  keys:
    TRITON_API_KEY: TRITON_API_KEY
capability:
  consumes:
  - type: http
    namespace: http-rest-inference
    baseUri: http://localhost:8000
    description: Triton Inference Server NVIDIA Triton Inference Server HTTP/REST API — Inference business capability. Self-contained,
      no shared references.
    resources:
    - name: v2-models-model_name-infer
      path: /v2/models/{model_name}/infer
      operations:
      - name: modelinfer
        method: POST
        description: Triton Inference Server Run Inference on a Model
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
        inputParameters:
        - name: body
          in: body
          type: object
          description: Request body (JSON).
          required: true
    - name: v2-models-model_name-versions-model_version-infer
      path: /v2/models/{model_name}/versions/{model_version}/infer
      operations:
      - name: modelversioninfer
        method: POST
        description: Triton Inference Server Run Inference on a Specific Model Version
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
        inputParameters:
        - name: body
          in: body
          type: object
          description: Request body (JSON).
          required: true
  exposes:
  - type: rest
    namespace: http-rest-inference-rest
    port: 8080
    description: REST adapter for Triton Inference Server NVIDIA Triton Inference Server HTTP/REST API — Inference. One Spectral-compliant
      resource per consumed operation, prefixed with /v1.
    resources:
    - path: /v1/v2/models/{model-name}/infer
      name: v2-models-model-name-infer
      description: REST surface for v2-models-model_name-infer.
      operations:
      - method: POST
        name: modelinfer
        description: Triton Inference Server Run Inference on a Model
        call: http-rest-inference.modelinfer
        with:
          body: rest.body
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/v2/models/{model-name}/versions/{model-version}/infer
      name: v2-models-model-name-versions-model-version-infer
      description: REST surface for v2-models-model_name-versions-model_version-infer.
      operations:
      - method: POST
        name: modelversioninfer
        description: Triton Inference Server Run Inference on a Specific Model Version
        call: http-rest-inference.modelversioninfer
        with:
          body: rest.body
        outputParameters:
        - type: object
          mapping: $.
  - type: mcp
    namespace: http-rest-inference-mcp
    port: 9090
    transport: http
    description: MCP adapter for Triton Inference Server NVIDIA Triton Inference Server HTTP/REST API — Inference. One tool
      per consumed operation, routed inline through this capability's consumes block.
    tools:
    - name: triton-inference-server-run-inference
      description: Triton Inference Server Run Inference on a Model
      hints:
        readOnly: false
        destructive: false
        idempotent: false
      call: http-rest-inference.modelinfer
      with:
        body: tools.body
      outputParameters:
      - type: object
        mapping: $.
    - name: triton-inference-server-run-inference-2
      description: Triton Inference Server Run Inference on a Specific Model Version
      hints:
        readOnly: false
        destructive: false
        idempotent: false
      call: http-rest-inference.modelversioninfer
      with:
        body: tools.body
      outputParameters:
      - type: object
        mapping: $.