Log inSubmit server
Replicate MCP Server logo

Replicate MCP Server

Runs Replicate models through MCP for image, video, audio, speech, text, vision, 3D, and workflow automation.

InstallRepositoryWebsite
Full README Compare alternatives Embed widget Claim this listing

Install Replicate MCP Server

Pick your client, then add it in one click or paste the config.

STDIONode.js
~/Library/Application Support/Claude/claude_desktop_config.json
claude_desktop_config.json
{
  "mcpServers": {
    "replicate-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "replicate-mcp-server"
      ],
      "env": {
        "REPLICATE_API_TOKEN": "YOUR_VALUE_HERE"
      }
    }
  }
}

Settings → Developer → Edit Config, merge this into mcpServers, then fully quit and reopen Claude Desktop.

Values are written into the snippet above so you can paste it as is. They never leave your browser.

Install not yet sandbox-tested— the server itself is confirmed live (we called its tools/list endpoint). The stdio install command is a separate, ongoing check.

Overview

replicate-mcp-server MCP server connects MCP clients to Replicate's model catalog through stdio or HTTP/SSE. It provides curated tools for generating and transforming media, running language and vision models, uploading files, discovering model schemas, and executing arbitrary Replicate models. Outputs can include Replicate URLs, downloaded local files, text results, or prediction IDs for later polling. Reach for it when an agent needs Replicate-hosted multimedia or AI models without writing separate API integrations.

Use cases

  • Generate images, videos, music, and speech from prompts
  • Run Replicate language and vision models from an MCP client
  • Build multi-step media workflows with prediction pipelines
  • Process multiple model requests concurrently with batch jobs
  • Upload local media and pass it into Replicate models

Key features

  • Curated multimedia generation tools
  • Generic Replicate model execution
  • Model search and schema inspection
  • Batch and DAG pipeline orchestration
  • Local output downloads
  • Prediction and training management

Tools36

What Replicate MCP Server lets your agent do. Expand a tool for its parameters.

Verified live~16.3k tokens of context

Security & health

What it can reach and how reliably it runs — from our own checks, not the maintainer’s claims.

ActiveLast checked 2d ago

Recent health check succeeded.

7/7 checks healthy over the last 46d

What it can do

From the 36 tools this server returned when we ran it; 36 declare their own safety hints.

Can delete or modify data · 5Writes files · 1Makes network requests · 32Read-only tools · 30
Network at startup3 connections
Files written at startupNone
Install scriptsNone
npm provenanceNot published
Package age136 days
Maintainers1

Observed 2026-10-16 in an isolated sandbox. A description of reach, not a security audit.

Supply-chain signal

No high-severity advisories surfaced by our automated scan.

Critical 0High 0Medium 0Low 0

Scanned 10d ago via OSV.dev · replicate-mcp-server (npm)

Adoption & maintenance

Factual signals from GitHub, npm, and our automated checks — not a rating.

GitHub stars4Stargazers on the source repository.
npm downloads276Package downloads in the last 30 days.
Last commit4d agoMost recent push to the default branch.
Tools exposed36Callable tools this server registers over MCP.
Directory activity3 viewsConfig copies, upvotes, and views on AllMCPs.

How Replicate MCP Server works

What replicate-mcp-server does

replicate-mcp-server MCP server exposes Replicate models as MCP tools for the 🎥 Multimedia Process category. Its curated capabilities cover image, video, music, speech, language, vision, upscaling, background removal, transcription, inpainting, segmentation, embeddings, voice cloning, 3D generation, and lip syncing. A generic model runner extends coverage to models that are not represented by a curated tool.

The server also includes model search, model-schema lookup, model recommendations, approximate cost estimates, prediction history, cancellation, file upload, asynchronous batches, prediction pipelines, and training-related operations. The supplied tool set is described as 36 tools, with curated model registries for common tasks.

How it works

Requests are sent to Replicate using the authenticated account's API token. Curated tools accept task-oriented inputs such as prompts, image URLs, audio URLs, model keys, aspect ratios, durations, seeds, and model-specific extra inputs. The generic runner accepts a full Replicate model identifier and an input object matching that model's schema.

Most generation tools wait for a result and can download output files locally. Images, video, audio, and other file outputs return Replicate URLs and local paths when downloading is enabled; text-oriented tools return content in text_output. Replicate URLs generally expire after about 24 hours. Long-running requests can return a prediction ID with a pending state, which can then be checked with replicate_get_prediction.

Batch jobs support up to 50 predictions, while pipelines connect dependent steps with references such as $stepId.urls[0]. Both workflows run in the background and expose status-polling tools.

Setup and configuration

The replicate-mcp-server MCP server can be installed locally with npx -y replicate-mcp-server, configured in an MCP client, or installed through the Smithery server page. Claude Desktop also supports the released .mcpb package. For remote use, the README provides an HTTP/SSE endpoint at https://replicate-mcp.sena-labs.dev/mcp; requests use the x-replicate-api-token header.

Set REPLICATE_API_TOKEN for local operation. The token authenticates calls to Replicate, so the account must have access to the models and operations being requested. Local and remote deployments use different token-handling arrangements; the remote endpoint expects the token in the request header.

The server supports stdio for local desktop and editor integrations, and HTTP/SSE for remote clients or custom applications. Generated files are downloaded under ~/Downloads/replicate-mcp/<model>/<prediction_id>/ by default when downloading is enabled.

Tools and capabilities

  • Generate images, videos, music, speech, 3D meshes, and lip-synced videos.
  • Analyze images, transcribe audio or video, create embeddings, upscale images, remove backgrounds, segment images, and inpaint masked areas.
  • Run Replicate-hosted language models and arbitrary catalog models.
  • Search models, inspect input and output schemas, recommend curated models, and estimate costs.
  • Upload local files or base64 data for use as model inputs.
  • Start and monitor concurrent batches, dependency-based pipelines, predictions, and training runs.

Limitations and notes

The replicate-mcp-server MCP server depends on Replicate availability, model-specific schemas, account permissions, and API billing. Curated shortcuts are not accepted by batch and pipeline starts; those operations require full owner/name identifiers. Model execution time varies considerably, especially for video, music, 3D, and training jobs.

Returned URLs and uploaded-file URLs expire after roughly 24 hours. Pipeline and batch state is held in server memory and can be lost if the server restarts. Cost estimates are approximate snapshots rather than billing guarantees. For image-producing tools, the client should use the embed markup returned by the tool when inline display is required.

Read the full README View source on GitHub

Reviews

No reviews yet — be the first to share how it worked for you.

Frequently Asked Questions about Replicate MCP Server

replicate-mcp-server is an MCP server that connects MCP clients to Replicate's model catalog. Its main tools generate and transform images, video, audio, speech, 3D assets, and text; analyze images; transcribe media; run arbitrary Replicate models; upload files; and manage predictions, batches, pipelines, and training jobs.

Install it locally with `npx -y replicate-mcp-server`, add it to an MCP client configuration, use the Smithery installation flow, or install the released `.mcpb` package in Claude Desktop. The README also provides a hosted HTTP/SSE endpoint for remote use.

It requires `REPLICATE_API_TOKEN` for local operation. The token authenticates requests to Replicate. When using the hosted HTTP/SSE endpoint, pass the token with the `x-replicate-api-token` request header instead.

Yes. The README names Cursor as a supported MCP client. It also names Claude Desktop, Cline, Continue in VS Code, Windsurf, the claude.ai web app through Connectors, and custom applications.

It can return a pending result with a prediction ID when a request exceeds its timeout. Call `replicate_get_prediction` with that ID to check progress and retrieve outputs after completion. Batch and pipeline jobs use their corresponding status tools.

Related MCP servers

All in Multimedia Process All alternatives
Add a badge to your READMEShow it’s listed — live SVG, Markdown or HTML.

AllMCPs Directory Badge

Full Badge Customizer

Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.

Badge Style:
Live Dynamic SVG PreviewReplicate MCP Server AllMCPs Directory Badge
Markdown (GitHub README)
[![AllMCPs](https://allmcps.com/api/badge/replicate-mcp-server?style=directory)](https://allmcps.com/mcp/replicate-mcp-server)
HTML Embed
<a href="https://allmcps.com/mcp/replicate-mcp-server"><img src="https://allmcps.com/api/badge/replicate-mcp-server?style=directory" alt="Replicate MCP Server on AllMCPs" /></a>