GenAI Custom Models

Generated on 29 Jun 2026

Copy page as Markdown View page as Markdown

This content is automatically generated from https://github.com/digitalocean-labs/mcp-digitalocean/blob/main/pkg/registry/genai-custom-models/README.md.

This package provides MCP tools for managing custom (bring-your-own) models on DigitalOcean’s GenAI platform.

Overview

The custom models tools enable users to:

  • List custom models with status filtering and pagination
  • Import models from HuggingFace or other sources
  • Update model metadata (name, description, tags, and — for Spaces-imported models — input/output modalities, parameter count, license)
  • Delete custom models

Tools

genai-custom-models-list

List custom models with optional status filter and pagination.

Arguments:

  • status (string, optional): Filter by status (STATUS_IMPORTING, STATUS_READY, STATUS_FAILED, STATUS_DELETED)
  • page (number, optional): Page number (default: 1)
  • per_page (number, optional): Results per page (default: 20)

Returns: JSON object with array of models and count

{
  "models": [
    {
      "uuid": "...",
      "name": "my-mistral-7b",
      "status": "STATUS_READY",
      "architecture": "MistralForCausalLM",
      "source_type": "SOURCE_TYPE_HUGGINGFACE"
    }
  ],
  "count": 1,
  "max_threshold": 10
}

genai-custom-models-import

Import a custom model from an external source (e.g. HuggingFace). Starts an async import job.

Consent (required every import): Present import terms and obtain explicit consent (yes) before calling. Set accept_terms_and_conditions to true only after the user agrees. Consent is validated before Hugging Face commit_sha resolution or the DigitalOcean API. This applies to every import, including re-imports of the same model. The tool rejects omitted or false values.

Arguments:

  • name (string, optional): Display name (leading/trailing spaces are trimmed when provided)
  • source_type (string, required): SOURCE_TYPE_HUGGINGFACE, SOURCE_TYPE_SPACES_BUCKET, SOURCE_TYPE_SDK_UPLOAD, SOURCE_TYPE_FINE_TUNING
  • source_ref (object, required): Source reference with repo_id, commit_sha (optional for HuggingFace; if omitted, fetched from Hugging Face Hub before the import API is called), access_type, hf_token (for private/gated)
  • accept_terms_and_conditions (boolean, required): Must be true after explicit user consent in chat
  • description (string, optional): Model description
  • preferred_gpu_region (string, optional): Preferred GPU region (e.g. nyc3)
  • tags (object, optional): Tags object with a tags array of strings

Returns: JSON object with model, import job status, and validation steps

{
  "model": { "uuid": "...", "name": "my-model", "status": "STATUS_IMPORTING" },
  "import_job": { "uuid": "...", "status": "...", "files_total": 12, "files_done": 0 },
  "validation_steps": [{ "name": "config_json", "passed": true }],
  "error": ""
}

genai-custom-models-update-metadata

Update the metadata of an existing custom model.

Arguments:

  • uuid (string, required): UUID of the custom model
  • name (string, optional): New name
  • description (string, optional): New description
  • tags (object, optional): New tags object with a tags array
  • input_modalities (array of strings, optional): Input modalities the model accepts (e.g. ["text", "image"]). Spaces-imported models only.
  • output_modalities (array of strings, optional): Output modalities the model produces (e.g. ["text"]). Spaces-imported models only.
  • parameters (string, optional): Parameter count as a numeric string (e.g. "7000000000" for 7B). Spaces-imported models only.
  • license (string, optional): License identifier (e.g. "apache-2.0"). Spaces-imported models only.

At least one updatable field must be provided.

Spaces-only capability fields. input_modalities, output_modalities, parameters, and license are auto-populated for HuggingFace imports and cannot be edited on those models — the backend rejects such requests with InvalidArgument. They are editable only for models with source_type = SOURCE_TYPE_SPACES_BUCKET, where the import path leaves them empty.

Modality semantics. When input_modalities or output_modalities is provided, the value replaces the existing list wholesale. Passing an empty array ([]) is rejected — a model must have at least one input and one output modality. To leave a modality list unchanged, simply omit the field.

Canonical modality vocabulary. The HuggingFace importer and existing catalog entries use this lowercase, singular set: text, image, audio, video. Prefer these tokens so Spaces-imported models stay aligned with HF-imported ones downstream.

Returns: JSON object with the updated model

genai-custom-models-delete

Delete a custom model by exact UUID or exact name.

Exact identifier: Provide uuid OR name (at least one). Only a full UUID or an exact model name deletes. Partial uuid or partial name returns candidates only — never deletes, even with a single match. Do not substitute values from a partial-match list.

Consent (required for every delete): Before calling with confirm_deletion: true, tell the user which model will be deleted (name and uuid), that deletion is permanent, and obtain explicit consent (yes).

Arguments:

  • name (string, optional): Exact custom model name (case-sensitive after trimming). Partial names return name matches only.
  • uuid (string, optional): Exact full model UUID. Partial uuids return uuid matches only.
  • confirm_deletion (boolean): Must be true after explicit user consent when performing a delete

Returns: JSON object with status on success

{
  "status": "DELETE_CUSTOM_MODEL_STATUS_SUCCESS",
  "error": ""
}

Returns (name not exact): JSON object listing candidate models

{
  "message": "no custom model with exact name \"llama\"; ask the user to provide the exact name ...",
  "query": "llama",
  "matches": [
    { "uuid": "...", "name": "my-llama-model", "status": "STATUS_READY" }
  ],
  "query_field": "name",
  "requires_exact_match": true,
  "do_not_substitute_name_or_uuid_from_matches": true
}

Custom Model Status Values

  • STATUS_IMPORTING: Model files are being imported
  • STATUS_READY: Model is ready for deployment
  • STATUS_FAILED: Import failed
  • STATUS_DELETED: Model has been deleted

Source Types

  • SOURCE_TYPE_HUGGINGFACE: Import from HuggingFace Hub
  • SOURCE_TYPE_SPACES_BUCKET: Import from DigitalOcean Spaces
  • SOURCE_TYPE_SDK_UPLOAD: Upload via SDK
  • SOURCE_TYPE_FINE_TUNING: Result of fine-tuning

Access Types (for source_ref)

  • ACCESS_TYPE_PUBLIC: Publicly accessible model
  • ACCESS_TYPE_PRIVATE: Private model (requires token)
  • ACCESS_TYPE_GATED: Gated model (requires acceptance + token)

Workflow Example 

1. Ask the user to accept import terms (storage cost, license, source). Wait for explicit yes.

2. Import a model from HuggingFace:
   genai-custom-models-import
     name: "my-mistral-7b"
     source_type: "SOURCE_TYPE_HUGGINGFACE"
     source_ref: { "repo_id": "mistralai/Mistral-7B-v0.1", "access_type": "ACCESS_TYPE_PUBLIC" }
     accept_terms_and_conditions: true

3. Check import status by listing models:
   genai-custom-models-list
     status: "STATUS_IMPORTING"

4. Once ready, update metadata:
   genai-custom-models-update-metadata
     uuid: "<uuid from step 2>"
     description: "Production model for customer support"
     tags: { "tags": ["production", "v1"] }

5. For a Spaces-imported model, also fill in the capability fields the importer
   could not infer:
   genai-custom-models-update-metadata
     uuid: "<spaces-model-uuid>"
     input_modalities: ["text", "image"]
     output_modalities: ["text"]
     parameters: "7000000000"
     license: "apache-2.0"

6. Confirm deletion with the user (permanent removal). Wait for explicit yes.

7. Delete using the exact name or uuid the user confirmed:
   genai-custom-models-delete
     name: "my-mistral-7b"
     confirm_deletion: true
   # or
   genai-custom-models-delete
     uuid: "<full-uuid>"
     confirm_deletion: true

We can't find any results for your search.

Try using different keywords or simplifying your search terms.