Replicate
Run thousands of AI models on demand: image generation, LLMs, video, audio, and more. Pay-per-use, no GPU management.
setup_integration or the Dashboard Vault — they're encrypted
with AES-256-GCM and never exposed.
Required Credentials
Get your API key at replicate.com →
Endpoints (3)
Run an AI model. Provide the model version and input parameters. Returns a prediction ID — poll with get-prediction for results.
| Field | Type | Required | Description |
|---|---|---|---|
version |
string | ✓ Yes | Model version hash. Find on replicate.com model pages. Example: "stability-ai/sdxl:..." for SDXL image gen. |
input |
object | ✓ Yes | Model input parameters. Varies per model. Example for SDXL: {"prompt": "A cat in space"}. |
Get prediction status and output. Poll this after create-prediction. Status: starting → processing → succeeded/failed.
| Field | Type | Required | Description |
|---|---|---|---|
prediction_id |
string | ✓ Yes | Prediction ID from create-prediction response. |
Search public models on Replicate. Returns model name, description, and run count.
No input parameters required.
MCP Tool Names
When using this integration through an AI assistant (Claude, ChatGPT, Cursor, etc.), the endpoints are available as MCP tools:
| Endpoint | MCP Tool Name |
|---|---|
| create-prediction | replicate_create_prediction |
| get-prediction | replicate_get_prediction |
| list-models | replicate_list_models |