Replicate MCP

How to Install and Use Replicate MCP

Replicate MCP is a simple and powerful tool that lets you connect natural language chat to Replicate’s API. This means you can search models, run models, and get results using easy chat commands. Here is a straightforward guide to get you started with Replicate MCP and run it on your computer or use it with apps like Claude or Cursor.

Installing the Local Replicate MCP Server

To start using the local Replicate MCP server, you need to run a command in your terminal or command prompt. The server runs using Node.js with one command that does not require prior installation.

Here’s what to do:

1. Open your terminal.
2. Run this command:

   npx -y replicate-mcp
   

   This will start the local MCP server on your machine. When you want to stop it, press Ctrl + c.

Although you can run the server this way, many apps can start it for you automatically by using a special JSON configuration file. Next, you'll learn how to set it up with popular apps.

Using Replicate MCP with Claude Desktop

Claude Desktop supports Replicate MCP out of the box, but only the desktop app version.

Here’s how to set it up simply:

1. Create a Replicate API token at replicate.com/account/api-tokens and copy it.
2. Open Claude Desktop.
3. Go to the Claude menu and select Settings… (not the in-app account settings).
4. Click Developer in the sidebar, then click Edit Config. This opens a file named claude_desktop_config.json.
5. Add this JSON to the file, replacing "your-token-here" with your token:

   {
       "mcpServers": {
           "replicate": {
               "command": "npx",
               "args": ["-y", "replicate-mcp"],
               "env": {
                   "REPLICATE_API_TOKEN": "your-token-here"
               }
           }
       }
   }
   

6. Restart Claude Desktop.
7. Click the Search and Tools icon in the chat input box. You should see replicate listed.

Now you are ready to chat naturally with models on Replicate using Claude Desktop.

Using Replicate MCP with Cursor

Cursor also supports Replicate MCP, letting you access Replicate models from your code editor.

You have two ways to set it up: quick or manual.

Quick Setup

Click the following link to install the MCP server for Cursor with your Replicate token:

Add replicate MCP server to Cursor

Make sure to replace the token before clicking if prompted.

Manual Setup

To set up manually:

1. Create a .cursor/mcp.json file in your project folder with this content (replace your token):

   {
       "mcpServers": {
           "replicate": {
               "command": "npx",
               "args": ["-y", "replicate-mcp"],
               "env": {
                   "REPLICATE_API_TOKEN": "your-token-here"
               }
           }
       }
   }
   

2. Open Cursor. The MCP server will start automatically or you can start it manually in the MCP settings.
3. Confirm replicate is listed under MCP settings.

You can now use Replicate tools directly inside Cursor via natural language commands.

Using Replicate MCP with Visual Studio Code and GitHub Copilot

You can also run Replicate MCP in Visual Studio Code with GitHub Copilot Chat.

Follow these steps:

1. Install VS Code 1.99 or newer and make sure Copilot Chat is enabled.
2. Create a Replicate API token here.
3. In your project, create a .vscode/mcp.json file with:

   {
       "servers": {
           "replicate": {
               "command": "npx",
               "args": ["-y", "replicate-mcp"],
               "env": {
                   "REPLICATE_API_TOKEN": "your-token-here"
               }
           }
       }
   }
   

4. Open this file in VS Code and click Start to launch the MCP server.
5. Open Copilot Chat, select Agent mode, and use Replicate commands like “Search Replicate for upscaler models.”

Summary

Replicate MCP makes it easy to interact with Replicate’s API using natural language. You can run it locally with a single command, add a JSON configuration to popular apps like Claude Desktop, Cursor, or VS Code, and start exploring models and running predictions quickly. Just remember to create and use your Replicate API token in the config files to authenticate the server.

This simple setup unlocks powerful AI model execution from chat or coding environments without complicated API calls.