Replicate MCP server for AI agents

Find models on Replicate, read their inputs, run them and fetch the output, and manage the account's predictions, deployments and trainings.

ReplicateBEARER_TOKEN

Runs with your users' own Replicate accounts. Credentials stay on prompt-tm.com.

23 tools · 0 triggers

Auth
Access token
Tools
23
Triggers
0
Category
AI
Connect Replicate replicate.com

Tools and triggers

  • Get the account

    REPLICATE_ACCOUNT_GET

    Returns the user or organization the API token belongs to: type, username, name and GitHub URL.read-only

  • Search models

    REPLICATE_SEARCH

    Searches public models, collections and docs by text, such as `upscale image` or `text to speech` (a beta endpoint).read-only

  • List public models

    REPLICATE_MODELS_LIST

    Lists public models, newest version first unless sorted otherwise. To find a model for a task, Search models is faster.read-only

  • Get a model

    REPLICATE_MODELS_GET

    Returns a model: description, run count, URLs and its latest version, whose `openapi_schema` describes the inputs and output.read-only

  • Get a model's README

    REPLICATE_MODELS_README_GET

    Returns a model's README as Markdown: what it does, its inputs and usage notes.read-only

  • List model examples

    REPLICATE_MODELS_EXAMPLES_LIST

    Lists example predictions the model's author saved, with their inputs and outputs. Useful to see what inputs work.read-only

  • List model versions

    REPLICATE_MODELS_VERSIONS_LIST

    Lists a model's versions, newest first, with their IDs and creation dates.read-only

  • Get a model version

    REPLICATE_MODELS_VERSIONS_GET

    Returns one version of a model with the `openapi_schema` of its inputs and output.read-only

  • List collections

    REPLICATE_COLLECTIONS_LIST

    Lists Replicate's curated model collections, such as super resolution or image restoration, with their slugs.read-only

  • Get a collection

    REPLICATE_COLLECTIONS_GET

    Returns a collection and the models in it.read-only

  • Run an official model

    REPLICATE_MODELS_PREDICTIONS_CREATE

    Runs an official model (one Replicate keeps running, such as black-forest-labs/flux-schnell) by owner and name, no version needed. For other models use Run a model version. Waits up to 20 seconds for the output; a slower prediction returns with status `starting` or `processing`, and Get a prediction fetches the result later. Running models is billed to the account.

  • Run a model version

    REPLICATE_PREDICTIONS_CREATE

    Runs a specific model version with the inputs you give. Waits up to 20 seconds for the output; a slower prediction returns with status `starting` or `processing`, and Get a prediction fetches the result later. Running models is billed to the account.

  • Run a deployment

    REPLICATE_DEPLOYMENTS_PREDICTIONS_CREATE

    Runs a prediction on one of the account's deployments (a model on dedicated hardware). Waits up to 20 seconds for the output; a slower prediction returns with status `starting` or `processing`, and Get a prediction fetches the result later. Running models is billed to the account.

  • Get a prediction

    REPLICATE_PREDICTIONS_GET

    Returns a prediction's status (`starting`, `processing`, `succeeded`, `failed`, `canceled`), its output, logs and error.read-only

  • List predictions

    REPLICATE_PREDICTIONS_LIST

    Lists the account's predictions, newest first, 100 per page.read-only

  • Cancel a prediction

    REPLICATE_PREDICTIONS_CANCEL

    Cancels a prediction that is still starting or running.

  • List deployments

    REPLICATE_DEPLOYMENTS_LIST

    Lists the account's deployments with the model, version and hardware of each one's current release.read-only

  • Get a deployment

    REPLICATE_DEPLOYMENTS_GET

    Returns a deployment and its current release: model, version, hardware and instance limits.read-only

  • List trainings

    REPLICATE_TRAININGS_LIST

    Lists the account's fine-tuning trainings, newest first.read-only

  • Get a training

    REPLICATE_TRAININGS_GET

    Returns a training's status, logs and, once it succeeded, the new model version it produced.read-only

  • Cancel a training

    REPLICATE_TRAININGS_CANCEL

    Cancels a training that is still running.

  • List hardware

    REPLICATE_HARDWARE_LIST

    Lists the hardware models and deployments can run on, with the SKU to pass when creating one.read-only

  • List files

    REPLICATE_FILES_LIST

    Lists the files the account uploaded for use as model inputs, with content type, size, checksums, expiry and URLs.read-only

Try Replicate now

Sign in, connect your account and run any of these in the playground, or ask your agent through the router.

Try it free

Connect via MCP

Create an MCP server for Replicate in the dashboard, then add its URL to your client. This is the Claude Code command; the client guide has the others.

terminal
claude mcp add --transport http replicate \
  "https://www.prompt-tm.com/mcp/<server id>?user_id=<your user>" \
  --header "x-api-key: ptm_…"

Connect via REST

Run a tool with one request. The response is { data, error, successful, log_id } for every tool.

terminal
curl -X POST https://www.prompt-tm.com/api/v3/tools/execute/REPLICATE_ACCOUNT_GET \
  -H "x-api-key: $PROMPT_TM_KEY" -H "content-type: application/json" \
  -d '{ "user_id": "alice", "arguments": {} }'

Replicate questions

Other toolkits