• SEARCH
  • INSPECT
  • EXECUTE
  • HUMAN

How to run machine learning models with your AI

Connect your AI client to Replicate once. From there, your agent handles the search, inspection, and execution of remote model predictions.

Ask AI about this page

Short answer

How can I run ML models through my AI?

Your agent uses this MCP to find models in the Replicate library and trigger predictions on their hardware. You get the model outputs directly in your chat. This breakdown shows how the agent handles the execution loop.

Execution outcomes

Where your requests land

The agent moves through these stages to complete a prediction.

SEARCH

Model discovery

The agent scans the Replicate library to find models that match your description. You get a list of available options to choose from.

INSPECT

Technical verification

The agent pulls the specific input schema for a model. This ensures the parameters you provide are correct before you run anything.

EXECUTE

Remote computation

The agent triggers the prediction on Replicate's hardware. You receive the resulting data or media once the task finishes.

HUMAN

Final review

You review the model outputs and decide if they meet your needs. The agent handles the technical execution, but you own the final result.

The workflow

What your AI does when a request arrives.

The agent manages the heavy lifting of interacting with Replicate's API.

  1. Locate models

    The agent searches the Replicate catalog to find models that fit your specific task.

    search_models

  2. Check requirements

    The agent retrieves the specific configuration and input details for a chosen model.

    get_model

  3. Trigger prediction

    The agent sends your input data to Replicate to start the remote run.

    create_prediction

  4. Fetch results

    The agent monitors the task and pulls the final output once the hardware finishes the job.

    get_prediction

Try it

Copy these to start.

Use these prompts to begin working with Replicate.

Starting points

These are opening lines. Swap out the model names or input descriptions for your specific needs.

Replicate API Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_ZjGhumvOdiFG6C4qZnstV0k6xajsQmuKnK85lhvV/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Replicate API capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "replicate-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_ZjGhumvOdiFG6C4qZnstV0k6xajsQmuKnK85lhvV/mcp"
    }
  }
}
Copy into chat04

  • Search Replicate for models that can upscale images.

  • Get the details for the stability-ai/sdxl model.

  • Run a prediction using the selected model with these inputs: [insert inputs].

  • Check the status of my last prediction to see if it's finished.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Replicate once, then ask.

One connection links your AI client to Replicate. Your credentials stay encrypted, and you can start running models immediately.

Connect Replicate to your AI

FAQ

How this task behaves

  • 01

    Can my AI delete models from Replicate?

    No. This MCP only allows searching, inspecting, and running predictions.

  • 02

    Does the AI pay for the model runs?

    No. The AI triggers the prediction, but the billing is handled through your own Replicate account.

  • 03

    Can I run models that aren't on Replicate?

    No. The agent is limited to the models available in the Replicate catalog.

  • 04

    Will the AI automatically run every model it finds?

    No. The agent finds and inspects models, but you must confirm the specific prediction you want to execute.

  • 05

    Can the AI change model settings?

    No. The agent can only use the parameters defined by the model's existing schema.

  • More questions about Replicate? The Connector page answers them. See everything the Replicate Connector can do

Connect Replicate API to Claude, Cursor, ChatGPT & more