Replicate MCP Server for Claude, VS Code, and Cursor

Safety notes

The server can invoke Replicate API capabilities through your account, so review prompts before running workflows that create predictions or consume credits.
Keep code-mode or execution-oriented workflows separate from simple model search unless you intend to allow generated code execution.

Privacy notes

Model prompts, input URLs, generated outputs, and API account metadata can flow through Replicate and the MCP client context.
Do not commit `REPLICATE_API_TOKEN` to a shared MCP config file.

Prerequisites

Node.js and npx available locally.
A Replicate API token stored outside version control.
Claude Desktop, Cursor, VS Code, or another MCP-compatible client.

Schema details

Install type: cli
Troubleshooting: No

Collection metadata

Estimated setup: 5 minutes
Difficulty: beginner

Full copyable content

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

About this resource

Content

Replicate MCP is Replicate's official MCP server for connecting AI clients to Replicate API tooling. The local server runs through npx -y replicate-mcp and uses a REPLICATE_API_TOKEN environment variable in client configuration. Replicate also publishes a hosted MCP surface for remote connection patterns.

Features

Local MCP server package launched with npx.
Works with Claude Desktop, Cursor, VS Code, and compatible MCP clients.
Lets agent workflows search Replicate and use Replicate API tooling from chat.
Supports token-backed account access through environment configuration.
Offers an advanced code-mode path for more complex SDK workflows.

Installation

Create a Replicate API token, then add the server to your MCP client config:

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

Restart the client and ask it to search Replicate for a small model category to confirm the tool is connected.

Use Cases

Find candidate Replicate models for image, audio, video, or language workflows.
Draft API calls using current model metadata.
Compare model inputs before building a workflow.
Let a coding assistant scaffold Replicate SDK usage with account-aware context.

Safety and Privacy

Treat the API token like any other paid service credential. Keep it in local MCP settings, a keychain-backed client, or an environment variable, and avoid sharing workspace-level config that contains the token. Review generated prompts and code before creating predictions that consume credits.

Duplicate Check

Existing tool listings include Replicate-adjacent AI tooling, but no dedicated MCP entry for the official Replicate server was found in content/mcp.