Replicate

You can run Replicate models through a lightweight MCP server that exposes a simple tool-based interface. This server lets you search models, run predictions, and handle images from Replicate directly in your MCP client. It’s useful for integrating Replicate’s capabilities into chat tools and automation workflows without leaving your client environment.

How to use

Connect your MCP client to the Replicate MCP server using the provided command and configuration. Once connected, you can browse models, run predictions with text or structured inputs, track and manage predictions, and view generated images in your browser. Use the available tools to search models, get detailed model information, create predictions, poll results, and manage image caches.

How to install

Prerequisites you need before starting are Node.js version 18 or newer and a working internet connection.

# Install the MCP server globally
npm install -g mcp-replicate

# Optional: run directly with npx without a global install
npx mcp-replicate

Additional sections

Configuration You must provide a Replicate API token for the server to operate. There are two ways to supply this token:

{
  "mcpServers": {
    "replicate": {
      "command": "mcp-replicate",
      "env": {
        "REPLICATE_API_TOKEN": "your_token_here"
      }
    }
  }
}

Environment variable option If you are using another MCP client, you can set the token as an environment variable in your shell before starting the server.

export REPLICATE_API_TOKEN=your_token_here

Troubleshooting

If tools are not visible or not functioning, verify the MCP client is configured with the correct server settings and that the Replicate API token is valid. Restart both the MCP client and the server if needed, and check for any error messages in the server output.

Notes on development and usage

This server is an experimental integration aimed at enabling tool-based access to Replicate through MCP clients. It provides a set of model, prediction, and image handling capabilities and can be used with a variety of MCP clients that support standard stdio-based servers.

Available tools

search_models

Find models using semantic search across available collections.

list_models

Browse the list of available models.

get_model

Retrieve detailed information about a specific model, including versions.

list_collections

Browse model collections to discover related assets.

get_collection

Get details about a specific collection, including its models.

create_prediction

Run a model with your inputs to generate a prediction.

create_and_poll_prediction

Run a prediction and wait until it completes, returning the result.

get_prediction

Check the status and result of a previously started prediction.

cancel_prediction

Cancel an ongoing prediction.

list_predictions

List your recent predictions for quick re-access.

view_image

Open a generated image in your browser.

clear_image_cache

Remove cached images to reclaim space or refresh visuals.

get_image_cache_stats

Check statistics for the image cache usage.