Skip to main content
Glama
Replicate Anywhere

Replicate Anywhere

Replicate Anywhere

An MCP (Model Context Protocol) server that gives AI assistants a way to generate and edit media: images, videos, looping clips, audio, music, speech, and 3D. It runs any model on Replicate.

The assistant reaches for it in three situations. The user names a model, such as Nano Banana, Kling 2.0, or Seedance. Or the user simply asks for an asset, such as "a video of a beach that loops", without mentioning Replicate or any model. Or a larger task, such as a polished presentation, website, or document, would look better with custom images or video. No hardcoded model lists: the AI searches Replicate for the right model and runs it.

Features

  • πŸ” Smart Model Search - Find models by fuzzy name matching (e.g., "flux", "stable diffusion", "nano banana pro")

  • πŸ€– Automatic Model Discovery - AI searches first, asks questions later

  • πŸ“‹ Parameter Detection - Automatically retrieves and understands model input schemas

  • πŸ–ΌοΈ Inline Image Display - Image outputs are formatted as markdown for inline display

  • ⏱️ Async Prediction Handling - Long-running predictions return status URLs instead of timing out

  • βœ… Prediction Status Checking - Check on running predictions that haven't completed yet

Related MCP server: OpenRouter MCP Server

Installation

Prerequisites

NPM (Global)

npm install -g replicate-anywhere

From Source

git clone https://github.com/fifthseason-ai/replicate-anywhere.git
cd replicate-anywhere
npm install
npm run build

Configuration

Environment Variables

Variable

Required

Default

Description

REPLICATE_API_TOKEN

Yes

-

Your Replicate API token

MAX_POLL_TIME

No

300000

Maximum time (ms) to wait for predictions before returning async status

MCP Client Configuration

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "replicate-anywhere": {
      "command": "npx",
      "args": ["-y", "replicate-anywhere"],
      "env": {
        "REPLICATE_API_TOKEN": "r8_your_token_here"
      }
    }
  }
}

LibreChat

Add to your librechat.yaml:

mcpServers:
  replicate-anywhere:
    type: stdio
    command: npx
    args:
      - -y
      - replicate-anywhere
    env:
      REPLICATE_API_TOKEN: "${REPLICATE_API_TOKEN}"

Docker

services:
  replicate-anywhere:
    build:
      context: ./replicate-anywhere
    environment:
      REPLICATE_API_TOKEN: ${REPLICATE_API_TOKEN}

Tools

search-models

Search for AI models on Replicate by name or description. This tool is designed to be called first when a user mentions any model name.

{
  "query": "flux pro"
}

get-model-info

Get detailed information about a specific model, including its input parameters schema.

{
  "owner": "black-forest-labs",
  "name": "flux-pro"
}

run-model

Run a prediction on any Replicate model.

{
  "model": "black-forest-labs/flux-pro",
  "input": {
    "prompt": "A beautiful sunset over mountains",
    "aspect_ratio": "16:9"
  }
}

list-models

List public models on Replicate (paginated).

{
  "cursor": "optional_pagination_cursor"
}

check-prediction

Check the status of a running prediction.

{
  "prediction_id": "abc123xyz"
}

Usage Examples

Generate an Image

User: "Generate an image of a cat wearing a space helmet using flux"

The AI will:

  1. Call search-models with query "flux"

  2. Call get-model-info to get parameters for the best match

  3. Call run-model with appropriate parameters

  4. Return the image inline (markdown formatted)

Use a Specific Model

User: "Use stable diffusion xl to create a cyberpunk cityscape"

The AI will search for "stable diffusion xl", find stability-ai/sdxl, and run it.

Check a Long-Running Prediction

User: "Check on my prediction abc123"

The AI will call check-prediction to get the current status and output if complete.

Output Formatting

Images

When a model returns image URLs, the output is automatically formatted as markdown:

**Generated Image:**

![Generated Image](https://replicate.delivery/...)

**Direct link:** https://replicate.delivery/...

Other Outputs

Non-image outputs are returned as JSON.

Development

# Install dependencies
npm install

# Build
npm run build

# Watch mode
npm run dev

# Run locally
REPLICATE_API_TOKEN=your_token npm start

Architecture

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”     β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚   AI Assistant  │────▢│ replicate-anywhere│────▢│  Replicate API  β”‚
β”‚  (Claude, etc.) │◀────│    MCP Server    │◀────│                 β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜     β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

The server acts as a bridge between MCP-compatible AI assistants and the Replicate API, providing:

  • Tool definitions that guide the AI on how to search and run models

  • Smart response formatting for different output types

  • Timeout handling for long-running predictions

License

MIT

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

Credits

Built by Aaron Sherrill

Available Tools

5 tools
check-predictionA

Check the status of a running prediction. Use this if a previous run-model call returned a prediction_id with status "processing".

Inputs: prediction_id (required).

get-model-infoA

Get detailed information about a specific Replicate model including its input parameters schema. Call this AFTER search-models to get parameter details for the best matching model before running it.

Inputs: name (required), owner (required).

list-modelsA

List public models on Replicate. Use search-models instead when the user mentions a specific model name - this tool is only for browsing all available models without a specific query.

Inputs: cursor.

run-modelA

Run a prediction on any Replicate model. WORKFLOW: If the user provides a fuzzy model name, first call search-models to find the exact model identifier, then call get-model-info to understand required parameters, then call this tool.

Inputs: input (required), model (required).

search-modelsA

IMPORTANT: This tool MUST be called FIRST before asking any clarifying questions about model names. When a user mentions ANY model name (exact, fuzzy, or incomplete like "flux", "stable diffusion", "nano banana pro"), immediately search for it. Do NOT ask the user to confirm model names - search first, then pick the best match from results based on name similarity and run count. Only ask for clarification if the search returns zero results.

Inputs: query (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.2/5.0

    Scored across 5 tools

    Disambiguation5/5

    Each tool has a distinct role in the model-running workflow: search vs browse, get info, run, and check status. There is no overlap between search-models and list-models because the former is for specific queries and the latter for browsing.

    Naming Consistency5/5

    All tool names use a consistent verb_noun pattern in snake_case: search-models, get-model-info, run-model, list-models, check-prediction. This makes the API predictable and easy to navigate.

    Tool Count5/5

    With 5 tools, the server is well-scoped for its purpose of running Replicate models. Each tool covers a necessary step in the workflow without unnecessary bloat.

    Completeness4/5

    The core lifecycle of searching, inspecting, running, and checking predictions is fully covered. A minor gap is the lack of a cancel-prediction tool, but this is not essential for the primary workflow.

    Maintenance

    ActivitySlowing
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      An MCP server that enables AI applications to access 20+ model providers (including OpenAI, Anthropic, Google) through a unified interface for text and image generation.
      2
      30
      MIT
    • F
      license
      A
      quality
      D
      maintenance
      An MCP server for discovering and querying over 300 AI models available on OpenRouter. It enables users to list, search, filter, compare, and get detailed information about models with pricing, context limits, and capabilities.
      5
      1
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      A plugin-based MCP server that enables AI assistants to interact with external systems through custom tools, resources, and prompts.
      4
      AGPL 3.0