Dedicated Inference
Generated on 28 Apr 2026
This content is automatically generated from https://github.com/digitalocean-labs/mcp-digitalocean/blob/main/pkg/registry/dedicated-inference/README.md.
Dedicated Inference (dedicated-inference)
What this service does
This package registers Dedicated Inference tools for the DigitalOcean MCP server. Dedicated Inference lets you run model workloads on GPU-backed infrastructure in your account. The tools call the public DigitalOcean API (/v2/dedicated-inferences) through [godo](https://github.com/digitalocean/godo).
Enable these tools by including the service key when you configure the server (see the main project docs): service name **dedicated-inference**.
Code layout
| Path | Purpose |
|---|---|
[dedicated_inference_tools.go](dedicated_inference_tools.go) |
Tool handlers and MCP tool definitions |
[dedicated_inference_tools_test.go](dedicated_inference_tools_test.go) |
Unit tests (mocked API client) |
[spec/](spec/) |
Optional JSON Schema snippets describing HTTP request bodies (reference only; not loaded at runtime) |
Tools exposed
| Tool | What it does |
|---|---|
dedicated-inference-create |
Create a new Dedicated Inference instance. Returns the resource and an optional initial auth token. |
dedicated-inference-get |
Fetch one instance by ID. |
dedicated-inference-list |
List instances with optional Region, Name, Page, and PerPage. Result is a JSON array only (no pagination metadata in the tool output). |
dedicated-inference-update |
Update an instance’s deployment spec and/or Hugging Face secret. |
dedicated-inference-delete |
Delete an instance by ID. |
Arguments (summary)
dedicated-inference-create
| Argument | Required | Description |
|---|---|---|
Name |
yes | Instance name |
Region |
yes | Region slug (e.g. nyc2) |
ModelDeployments |
yes | Non-empty list of deployments (ModelSlug, ModelProvider, optional ModelID, Accelerators with AcceleratorSlug, Scale, optional Type) |
EnablePublicEndpoint |
no | Public endpoint toggle |
VPCUUID |
no | VPC UUID; omit to use platform defaults |
HuggingFaceToken |
no | Write-only; not returned in responses |
The handler sets spec.version to 1 on create.
dedicated-inference-update
| Argument | Required | Description |
|---|---|---|
DedicatedInferenceID |
yes | Instance UUID |
| Other fields | no | Same family as create (Name, Region, EnablePublicEndpoint, VPCUUID, ModelDeployments, HuggingFaceToken)—only values you pass are forwarded into the API request. For PATCH semantics and required fields, rely on the public API and the behavior of your request body. |
HuggingFaceToken |
no | Omit to leave existing secret; set to replace |
dedicated-inference-get / dedicated-inference-delete
- get:
DedicatedInferenceID(required). - delete:
DedicatedInferenceID(required). Returns a small JSON success payload.
dedicated-inference-list
- Optional:
Region,Name,Page,PerPage.
Reference schemas (spec/*.json)
| File | Contents |
|---|---|
[spec/dedicated-inference-create-schema.json](spec/dedicated-inference-create-schema.json) |
HTTP-style spec + optional secrets (snake_case) |
[spec/dedicated-inference-update-schema.json](spec/dedicated-inference-update-schema.json) |
Update body shape; MCP passes DedicatedInferenceID as a separate tool argument |
Secrets
HuggingFaceTokenis write-only in outputs.- On update, omitting it keeps the previous secret; setting it replaces the value.
Responses and polling
- Create returns
dedicated_inferenceplus optionaltoken(API-issued auth token, if returned). - Get / update return the dedicated inference resource JSON.
- Operations may complete asynchronously; use get to poll until status stabilizes.
Auth
Callers need a DigitalOcean API token with the appropriate **dedicated_inference** scopes for the operations they use.