Replicate · Capability
Replicate Model Inference

Workflow capability for running AI model inference on Replicate. Covers creating predictions against versioned models and production deployments, monitoring prediction status, and canceling running jobs. Designed for developers integrating AI generation into applications.
Run with Naftiko Artificial IntelligenceMachine LearningModel InferencePredictionsDeployments
What You Can Do

GET
List predictions — List all predictions.
/v1/predictions
POST
Create prediction — Create a new inference prediction.
/v1/predictions
GET
Get prediction — Get prediction status and output.
/v1/predictions/{id}
DELETE
Cancel prediction — Cancel a running prediction.
/v1/predictions/{id}
POST
Create model prediction — Run inference on an official Replicate model.
/v1/models/{owner}/{name}/predictions
POST
Create deployment prediction — Run inference against a production deployment.
/v1/deployments/{owner}/{name}/predictions
GET
List hardware — List available hardware for running models.
/v1/hardware
GET
List collections — List all curated model collections.
/v1/collections
GET
Get collection — Get a specific model collection.
/v1/collections/{slug}
MCP Tools

list-predictions

List all inference predictions for the account.
read-only
create-prediction

Run a new AI model inference prediction by specifying a model version and input.
get-prediction

Get the status and output of a prediction.
read-only
cancel-prediction

Cancel a currently running prediction.
idempotent
create-model-prediction

Run inference on an official Replicate model (no version required).
create-deployment-prediction

Run inference against a production deployment for lower latency.
list-hardware

List available GPU hardware options for running models.
read-only
list-collections

Browse curated model collections by category.
read-only
get-collection

Get all models in a specific curated collection.
read-only
Capability Spec

naftiko: 1.0.0-alpha2
info:
  label: Replicate Model Inference
  description: Workflow capability for running AI model inference on Replicate. Covers creating predictions against versioned
    models and production deployments, monitoring prediction status, and canceling running jobs. Designed for developers integrating
    AI generation into applications.
  tags:
  - Artificial Intelligence
  - Machine Learning
  - Model Inference
  - Predictions
  - Deployments
  created: '2026-05-02'
  modified: '2026-05-06'
binds:
- namespace: env
  keys:
    REPLICATE_API_TOKEN: REPLICATE_API_TOKEN
capability:
  consumes:
  - type: http
    namespace: replicate
    baseUri: https://api.replicate.com/v1
    description: Replicate REST API for running ML models and managing resources.
    authentication:
      type: bearer
      token: '{{REPLICATE_API_TOKEN}}'
    resources:
    - name: account
      path: /account
      description: Account information for the authenticated user or organization.
      operations:
      - name: get-account
        method: GET
        description: Get the authenticated account information.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: collections
      path: /collections
      description: Curated collections of models.
      operations:
      - name: list-collections
        method: GET
        description: List all collections of models.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: get-collection
        method: GET
        description: Get a specific collection of models by slug.
        inputParameters:
        - name: collection_slug
          in: path
          type: string
          required: true
          description: The slug of the collection.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: deployments
      path: /deployments
      description: Manage model deployments for production use.
      operations:
      - name: list-deployments
        method: GET
        description: List all deployments for the authenticated account.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: create-deployment
        method: POST
        description: Create a new model deployment.
        body:
          type: json
          data:
            name: '{{tools.name}}'
            model: '{{tools.model}}'
            version: '{{tools.version}}'
            hardware: '{{tools.hardware}}'
            min_instances: '{{tools.min_instances}}'
            max_instances: '{{tools.max_instances}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: get-deployment
        method: GET
        description: Get a specific deployment by owner and name.
        inputParameters:
        - name: deployment_owner
          in: path
          type: string
          required: true
          description: The deployment owner.
        - name: deployment_name
          in: path
          type: string
          required: true
          description: The deployment name.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: update-deployment
        method: PATCH
        description: Update a deployment configuration.
        inputParameters:
        - name: deployment_owner
          in: path
          type: string
          required: true
          description: The deployment owner.
        - name: deployment_name
          in: path
          type: string
          required: true
          description: The deployment name.
        body:
          type: json
          data:
            hardware: '{{tools.hardware}}'
            min_instances: '{{tools.min_instances}}'
            max_instances: '{{tools.max_instances}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: delete-deployment
        method: DELETE
        description: Delete a deployment.
        inputParameters:
        - name: deployment_owner
          in: path
          type: string
          required: true
          description: The deployment owner.
        - name: deployment_name
          in: path
          type: string
          required: true
          description: The deployment name.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: create-deployment-prediction
        method: POST
        description: Create a prediction using a deployment.
        inputParameters:
        - name: deployment_owner
          in: path
          type: string
          required: true
          description: The deployment owner.
        - name: deployment_name
          in: path
          type: string
          required: true
          description: The deployment name.
        body:
          type: json
          data:
            input: '{{tools.input}}'
            webhook: '{{tools.webhook}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: hardware
      path: /hardware
      description: Available hardware options for running models.
      operations:
      - name: list-hardware
        method: GET
        description: List all available hardware options.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: models
      path: /models
      description: ML models hosted on Replicate.
      operations:
      - name: list-models
        method: GET
        description: List all public models.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: create-model
        method: POST
        description: Create a new model.
        body:
          type: json
          data:
            owner: '{{tools.owner}}'
            name: '{{tools.name}}'
            description: '{{tools.description}}'
            visibility: '{{tools.visibility}}'
            hardware: '{{tools.hardware}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: search-models
        method: GET
        description: Search public models.
        inputParameters:
        - name: query
          in: query
          type: string
          required: false
          description: Search query string.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: get-model
        method: GET
        description: Get a specific model.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: delete-model
        method: DELETE
        description: Delete a model.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: create-model-prediction
        method: POST
        description: Create a prediction using an official model.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        body:
          type: json
          data:
            input: '{{tools.input}}'
            webhook: '{{tools.webhook}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: list-model-versions
        method: GET
        description: List all versions of a model.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: get-model-version
        method: GET
        description: Get a specific model version.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        - name: version_id
          in: path
          type: string
          required: true
          description: The version ID.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: delete-model-version
        method: DELETE
        description: Delete a specific model version.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        - name: version_id
          in: path
          type: string
          required: true
          description: The version ID.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: predictions
      path: /predictions
      description: Inference runs against ML models.
      operations:
      - name: list-predictions
        method: GET
        description: List all predictions for the authenticated account.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: create-prediction
        method: POST
        description: Create a new prediction.
        body:
          type: json
          data:
            version: '{{tools.version}}'
            input: '{{tools.input}}'
            webhook: '{{tools.webhook}}'
            webhook_events_filter: '{{tools.webhook_events_filter}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: get-prediction
        method: GET
        description: Get a specific prediction.
        inputParameters:
        - name: prediction_id
          in: path
          type: string
          required: true
          description: The prediction ID.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: cancel-prediction
        method: POST
        description: Cancel a running prediction.
        inputParameters:
        - name: prediction_id
          in: path
          type: string
          required: true
          description: The prediction ID.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: trainings
      path: /trainings
      description: Fine-tuning jobs for ML models.
      operations:
      - name: list-trainings
        method: GET
        description: List all trainings for the authenticated account.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: create-training
        method: POST
        description: Create a new training job on a model version.
        inputParameters:
        - name: model_owner
          in: path
          type: string
          required: true
          description: The model owner.
        - name: model_name
          in: path
          type: string
          required: true
          description: The model name.
        - name: version_id
          in: path
          type: string
          required: true
          description: The version ID to train.
        body:
          type: json
          data:
            destination: '{{tools.destination}}'
            input: '{{tools.input}}'
            webhook: '{{tools.webhook}}'
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: get-training
        method: GET
        description: Get a specific training job.
        inputParameters:
        - name: training_id
          in: path
          type: string
          required: true
          description: The training ID.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
      - name: cancel-training
        method: POST
        description: Cancel a running training job.
        inputParameters:
        - name: training_id
          in: path
          type: string
          required: true
          description: The training ID.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
    - name: webhooks
      path: /webhooks/default/secret
      description: Webhook signing secret management.
      operations:
      - name: get-webhook-secret
        method: GET
        description: Get the signing secret for the default webhook.
        outputRawFormat: json
        outputParameters:
        - name: result
          type: object
          value: $.
  exposes:
  - type: rest
    port: 8080
    namespace: replicate-inference-api
    description: Unified REST API for running ML model inference on Replicate.
    resources:
    - path: /v1/predictions
      name: predictions
      description: Create and manage model inference predictions.
      operations:
      - method: GET
        name: list-predictions
        description: List all predictions.
        call: replicate.list-predictions
        outputParameters:
        - type: object
          mapping: $.
      - method: POST
        name: create-prediction
        description: Create a new inference prediction.
        call: replicate.create-prediction
        with:
          version: rest.version
          input: rest.input
          webhook: rest.webhook
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/predictions/{id}
      name: prediction
      description: Get or cancel a specific prediction.
      operations:
      - method: GET
        name: get-prediction
        description: Get prediction status and output.
        call: replicate.get-prediction
        with:
          prediction_id: rest.id
        outputParameters:
        - type: object
          mapping: $.
      - method: DELETE
        name: cancel-prediction
        description: Cancel a running prediction.
        call: replicate.cancel-prediction
        with:
          prediction_id: rest.id
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/models/{owner}/{name}/predictions
      name: model-predictions
      description: Create predictions using official models.
      operations:
      - method: POST
        name: create-model-prediction
        description: Run inference on an official Replicate model.
        call: replicate.create-model-prediction
        with:
          model_owner: rest.owner
          model_name: rest.name
          input: rest.input
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/deployments/{owner}/{name}/predictions
      name: deployment-predictions
      description: Create predictions using production deployments.
      operations:
      - method: POST
        name: create-deployment-prediction
        description: Run inference against a production deployment.
        call: replicate.create-deployment-prediction
        with:
          deployment_owner: rest.owner
          deployment_name: rest.name
          input: rest.input
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/hardware
      name: hardware
      description: Available GPU hardware options.
      operations:
      - method: GET
        name: list-hardware
        description: List available hardware for running models.
        call: replicate.list-hardware
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/collections
      name: collections
      description: Curated model collections.
      operations:
      - method: GET
        name: list-collections
        description: List all curated model collections.
        call: replicate.list-collections
        outputParameters:
        - type: object
          mapping: $.
    - path: /v1/collections/{slug}
      name: collection
      description: Specific model collection.
      operations:
      - method: GET
        name: get-collection
        description: Get a specific model collection.
        call: replicate.get-collection
        with:
          collection_slug: rest.slug
        outputParameters:
        - type: object
          mapping: $.
  - type: mcp
    port: 9090
    namespace: replicate-inference-mcp
    transport: http
    description: MCP server for AI-assisted model inference on Replicate.
    tools:
    - name: list-predictions
      description: List all inference predictions for the account.
      hints:
        readOnly: true
        openWorld: false
      call: replicate.list-predictions
      outputParameters:
      - type: object
        mapping: $.
    - name: create-prediction
      description: Run a new AI model inference prediction by specifying a model version and input.
      hints:
        readOnly: false
        destructive: false
        idempotent: false
      call: replicate.create-prediction
      with:
        version: tools.version
        input: tools.input
        webhook: tools.webhook
      outputParameters:
      - type: object
        mapping: $.
    - name: get-prediction
      description: Get the status and output of a prediction.
      hints:
        readOnly: true
        openWorld: false
      call: replicate.get-prediction
      with:
        prediction_id: tools.prediction_id
      outputParameters:
      - type: object
        mapping: $.
    - name: cancel-prediction
      description: Cancel a currently running prediction.
      hints:
        readOnly: false
        destructive: true
        idempotent: true
      call: replicate.cancel-prediction
      with:
        prediction_id: tools.prediction_id
      outputParameters:
      - type: object
        mapping: $.
    - name: create-model-prediction
      description: Run inference on an official Replicate model (no version required).
      hints:
        readOnly: false
        destructive: false
        idempotent: false
      call: replicate.create-model-prediction
      with:
        model_owner: tools.model_owner
        model_name: tools.model_name
        input: tools.input
      outputParameters:
      - type: object
        mapping: $.
    - name: create-deployment-prediction
      description: Run inference against a production deployment for lower latency.
      hints:
        readOnly: false
        destructive: false
        idempotent: false
      call: replicate.create-deployment-prediction
      with:
        deployment_owner: tools.deployment_owner
        deployment_name: tools.deployment_name
        input: tools.input
      outputParameters:
      - type: object
        mapping: $.
    - name: list-hardware
      description: List available GPU hardware options for running models.
      hints:
        readOnly: true
        openWorld: false
      call: replicate.list-hardware
      outputParameters:
      - type: object
        mapping: $.
    - name: list-collections
      description: Browse curated model collections by category.
      hints:
        readOnly: true
        openWorld: true
      call: replicate.list-collections
      outputParameters:
      - type: object
        mapping: $.
    - name: get-collection
      description: Get all models in a specific curated collection.
      hints:
        readOnly: true
        openWorld: true
      call: replicate.get-collection
      with:
        collection_slug: tools.collection_slug
      outputParameters:
      - type: object
        mapping: $.