electrical_services

Replicate MCP Server — Hosted for Any AI Agent

MCP Server language Hosted language Public

Run any AI model on Replicate — image generation, video, audio, language models — and manage predictions, deployments, and your model library.

@aerostack verified

• v0.1.0 • MIT • Updated Jun 29, 2026

add_circleAdd to Workspace

robot_2

Use with AI AssistantsMCP

Connect Claude, Cursor, or any MCP-compatible client — then call tools directly

① Add This MCP Server

Paste into your AI client config — then all its tools are available instantly.

.claude/mcp.json

{
  "mcpServers": {
    "replicate": {
      "url": "https://mcp.aerostack.dev/s/aerostack/mcp-replicate",
      "headers": {
        "Authorization": "Bearer YOUR_AEROSTACK_TOKEN"
      }
    }
  }
}

Replace YOUR_AEROSTACK_TOKEN with your API token from the dashboard.

② Call a Tool

Ask your AI assistant to call a specific tool, or send raw JSON-RPC:

+7 more

Natural Language Prompt

“Use the _ping tool to verify replicate credentials by calling a lightweight read endpoint. used internally by aerostack to validate credentials”

Using a Workspace?

Add this MCP to your Workspace — your team shares one token, secrets are stored securely, and every AI agent in the workspace can call it without per-user setup.

add_circleAdd to Workspace

description Overview

mcp-replicate — Replicate MCP Server

Run any AI model on Replicate — image generation, video, audio, language models — and manage predictions, deployments, and your model library.

Live endpoint: https://mcp.aerostack.dev/s/aerostack/mcp-replicate

What You Can Do

This MCP server gives AI agents access to Replicate via 12 tools. Connect it to any Aerostack workspace and your agents can interact with Replicate directly.

Available Tools

Tool	Description
`run_model`	Run a Replicate model with a specific version and inputs. Returns prediction output or a prediction ID for async polling
`get_prediction`	Get the current status and output of a prediction by its ID
`cancel_prediction`	Cancel a prediction that is currently queued or in progress
`list_predictions`	List your recent predictions with status, model, and output URLs
`get_model`	Get details about a Replicate model: description, visibility, run count, and latest version
`list_model_versions`	List all available versions of a Replicate model with their creation dates and OpenAPI schemas
`get_model_version`	Get the OpenAPI input/output schema for a specific model version
`search_models`	Search Replicate public models by keyword, returning name, description, run count, and latest version
`list_deployments`	List your Replicate deployments (dedicated hosted model instances)
`create_deployment_prediction`	Run a prediction on a specific named deployment (useful for consistent latency with dedicated compute)
`get_account`	Get your Replicate account information: username, name, and account type
`create_model`	Create a new model on Replicate with a specified owner, name, visibility, and hardware

Configuration

Variable	Required	Description
`REPLICATE_API_TOKEN`	Yes	Your Replicate API token — found at replicate.com/account/api-tokens

Quick Start

Add to Aerostack Workspace

Go to aerostack.dev → Your Project → MCPs
Search for "Replicate" and click Add to Workspace

Add the following secrets under Project → Secrets:

REPLICATE_API_TOKEN

Once added, every AI agent in your workspace can use Replicate tools automatically.

Direct API Call

curl -X POST https://mcp.aerostack.dev/s/aerostack/mcp-replicate \
  -H 'Content-Type: application/json' \
  -H 'X-Mcp-Secret-REPLICATE-API-TOKEN: your-replicate-api-token' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"run_model","arguments":{}}}'

License

MIT

terminal Tools (13)

Available tools on this MCP server. Each tool can be called directly from any AI agent.

terminal

_ping #1

Verify Replicate credentials by calling a lightweight read endpoint. Used internally by Aerostack to validate credentials.

terminal

run_model #2

Run a Replicate model with a specific version and inputs. Returns prediction output or a prediction ID for async polling

terminal

get_prediction #3

Get the current status and output of a prediction by its ID

terminal

cancel_prediction #4

Cancel a prediction that is currently queued or in progress

terminal

list_predictions #5

List your recent predictions with status, model, and output URLs

terminal

get_model #6

Get details about a Replicate model: description, visibility, run count, and latest version

terminal

list_model_versions #7

List all available versions of a Replicate model with their creation dates and OpenAPI schemas

terminal

get_model_version #8

Get the OpenAPI input/output schema for a specific model version

terminal

search_models #9

Search Replicate public models by keyword, returning name, description, run count, and latest version

terminal

list_deployments #10

List your Replicate deployments (dedicated hosted model instances)

terminal

create_deployment_prediction #11

Run a prediction on a specific named deployment (useful for consistent latency with dedicated compute)

terminal

get_account #12

Get your Replicate account information: username, name, and account type

terminal

create_model #13

Create a new model on Replicate with a specified owner, name, visibility, and hardware

Details

upgrade Version 0.1.0

gavel License MIT

wifi Transport streamable-http

lock Access Public

category Category API Connectors

terminal Tools 13

language Live Endpoint

https://mcp.aerostack.dev/s/aerostack/mcp-replicate

Sub-50ms globally · Zero cold start

Publisher

@aerostack verified

Pre-built functions for the most common MCP tool patterns. Clone, extend, and deploy.

Frequently asked questions

What is the Replicate MCP server and what can it do? +

The Replicate MCP server is hosted on Aerostack and exposes these tools to your AI agent: `_ping`, `run_model`, `get_prediction`, `cancel_prediction`, `list_predictions`. You get one hosted URL — no self-hosting — that works from Claude, Cursor, ChatGPT, Gemini, VS Code, or any MCP-compatible client, and you can share it with your team or combine it with other MCP servers in a workspace.

Is the Replicate MCP server hosted, or do I have to run it myself? +

It's hosted on Aerostack's edge infrastructure — you don't deploy or maintain anything. Add it to a workspace and you get one authenticated URL, with secrets encrypted, that any AI agent or editor can connect to. Use it solo or share the same URL across your whole team.

Which AI agents and editors can use the Replicate MCP server? +

Any MCP client: Claude and Claude Code, Cursor, ChatGPT, Gemini, Windsurf, Cline, VS Code, and custom agents. Because it's one hosted URL, the same Replicate MCP server works everywhere — and you can compose it with other MCP servers, skills, and functions behind a single workspace URL.

How do I install the Replicate MCP server in Claude Desktop? +

Add the following to your Claude Desktop config (`claude_desktop_config.json`): ```json { "mcpServers": { "@aerostack/mcp-replicate": { "command": "npx", "args": ["-y", "@aerostack/@aerostack/mcp-replicate"] } } } ``` Then restart Claude Desktop and the tools will appear automatically.

How do I use the Replicate MCP server in Cursor? +

In Cursor, open **Settings → MCP** and add: ```json { "name": "@aerostack/mcp-replicate", "command": "npx", "args": ["-y", "@aerostack/@aerostack/mcp-replicate"] } ``` Save and reload Cursor. The MCP tools will be available in Agent mode.

Does Replicate MCP require authentication? +

Yes. Replicate requires authentication. Check the MCP's documentation for the required credentials.

Replicate MCP Server — Hosted for Any AI Agent

description Overview

mcp-replicate — Replicate MCP Server

What You Can Do

Available Tools

Configuration

Quick Start

Add to Aerostack Workspace

Direct API Call

License

terminal Tools (13)

Details

Publisher

Tags

More in API Connectors

Aerostack Registry

Algolia

Arangodb

Ayrshare

Basecamp

Bigquery

Frequently asked questions