MCPollinations Multimodal MCP Server

A Model Context Protocol (MCP) server that enables AI assistants to generate images, text, and audio through the Pollinations APIs

Features

• Generate image URLs from text prompts
• Generate images and return them as base64-encoded data AND save as png, jpeg, jpg, or webp (default: png)
• Generate text responses from text prompts
• Generate audio responses from text prompts
• List available image and text generation models
• No authentication required
• Simple and lightweight
• Compatible with the Model Context Protocol (MCP)

System Requirements

• Node.js: Version 14.0.0 or higher
• For best performance, we recommend Node.js 16.0.0 or higher
• Node.js versions below 16 use an AbortController polyfill

Quick Start

Installing via Smithery

To install mcpollinations for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @pinkpixel-dev/mcpollinations --client claude

The easiest way to use the MCP server:

# Run directly with npx (no installation required)
npx @pinkpixel/mcpollinations

If you prefer to install it globally:

# Install globally
npm install -g @pinkpixel/mcpollinations

# Run the server
mcpollinations
# or
npx @pinkpixel/mcpollinations

Or clone the repository:

# Clone the git repository
git clone https://github.com/pinkpixel-dev/mcpollinations.git
# Run the server
mcpollinations
# or
npx @pinkpixel/mcpollinations
# or run directly
node /path/to/MCPollinations/pollinations-mcp-server.js

MCP Integration

To integrate the server with applications that support the Model Context Protocol (MCP):

1. Generate an MCP configuration file:

# If installed globally
npx @pinkpixel/mcpollinations generate-config

# Or run directly
node /path/to/MCPollinations/generate-mcp-config.js

Quick MCP Config (env)

If you prefer to skip the generator, copy this into your MCP client config:

{
  "mcpollinations": {
    "command": "npx",
    "args": ["-y", "@pinkpixel/mcpollinations"],
    "env": {
      "token": "YOUR_TOKEN_OPTIONAL",
      "referrer": "your-app-or-domain-optional",
      "IMAGE_MODEL": "flux",
      "IMAGE_WIDTH": "1024",
      "IMAGE_HEIGHT": "1024",
      "IMAGE_ENHANCE": "true",
      "IMAGE_SAFE": "false",
      "TEXT_MODEL": "openai",
      "TEXT_TEMPERATURE": "0.7",
      "TEXT_TOP_P": "0.9",
      "TEXT_SYSTEM": "",
      "AUDIO_VOICE": "alloy",
      "OUTPUT_DIR": "./mcpollinations-output"
    }
  }
}

2. Follow the prompts to customize your configuration or use the defaults.

• Set an output directory (relative paths recommended for portability)
• Windows users: Consider using absolute paths (e.g., C:\Users\YourName\Pictures\MCPollinations) for more reliable file saving
• Configure optional authentication (token, referrer) under env
• Configure default parameters for image generation (with a list of available models, dimensions, etc.)
• Configure default parameters for text generation (with a list of available models)
• Configure default parameters for audio generation (voice)

3. Copy the generated mcp.json file to your application's MCP settings .json file.

4. Restart your application.

After integration, you can use commands like:

"Generate an image of a sunset over the ocean using MCPollinations"

Authentication (Optional)

MCPollinations supports optional authentication to provide access to more models and better rate limits. The server works perfectly without authentication (free tier), but users with API tokens can get enhanced access.

Configuration Methods

Method 1: Environment Variables (Recommended for security)

# Set environment variables before running the server
export POLLINATIONS_TOKEN="your-api-token"
export POLLINATIONS_REFERRER="https://your-domain.com"

# Then run the server
npx @pinkpixel/mcpollinations

Method 2: MCP Configuration File (env)

When generating your MCP configuration, place auth inside env so your MCP client passes them as environment variables to the server process:

{
  "mcpollinations": {
    "command": "npx",
    "args": ["-y", "@pinkpixel/mcpollinations"],
    "env": {
      "token": "your-api-token",
      "referrer": "your-app-or-domain"
    }
  }
}

You can also provide POLLINATIONS_TOKEN and POLLINATIONS_REFERRER instead; the server recognizes both forms. Using token and referrer inside env is recommended for MCP configs.

Authentication Parameters

• **token** (optional): Your Pollinations API token for enhanced access
• **referrer** (optional): Your domain/application referrer URL

Both parameters are completely optional. Leave them empty or unset to use the free tier.

Using Your Configuration Settings

MCPollinations respects your MCP configuration settings placed in env as defaults. When you ask an AI assistant to generate content:

• Your configured models, output directories, and parameters are used automatically
• To override: Specifically instruct the AI to use different settings
• "Generate an image using the kontext model"
• "Save this image to my Desktop folder"
• "Use a temperature of 1.2 for this text generation"

Example Instructions:

• ✅ "Generate a sunset image" → Uses your configured model and output directory
• ✅ "Generate a sunset image with the flux model" → Overrides model only
• ✅ "Generate a sunset image and save it to C:\Pictures" → Overrides output path only

This ensures your preferences are always respected unless you specifically want different settings for a particular request.

Troubleshooting

"AbortController is not defined" Error

If you encounter this error when running the MCP server:

ReferenceError: AbortController is not defined

This is usually caused by running on an older version of Node.js (below version 16.0.0). Try one of these solutions:

1. Update Node.js (recommended):

• Update to Node.js 16.0.0 or newer

2. Use Global Installation

• Update to the latest version of the package:

npm install -g @pinkpixel/mcpollinations
   # Run with npx
   npx @pinkpixel/mcpollinations

3. Install AbortController manually:

• If for some reason the polyfill doesn't work:

npm install node-abort-controller

Check Your Node.js Version

To check your current Node.js version:

node --version

If it shows a version lower than 16.0.0, consider upgrading for best compatibility.

Available Tools

The MCP server provides the following tools:

Image Generation Tools

1. generateImageUrl - Generates an image URL from a text prompt

2. generateImage - Generates an image, returns it as base64-encoded data, and saves it to a file by default (PNG format)

3. editImage - NEW! Edit or modify existing images based on text prompts

4. generateImageFromReference - NEW! Generate new images using existing images as reference

5. listImageModels - Lists available models for image generation

Text & Audio Tools

6. respondText - Responds with text to a prompt using text models (customizable parameters)

7. respondAudio - Generates an audio response to a text prompt (customizable voice parameter)

8. listTextModels - Lists available models for text generation

9. listAudioVoices - Lists all available voices for audio generation

Text Generation Details

Available Parameters

The respondText tool supports several parameters for fine-tuning text generation:

• **model**: Choose from available text models (use listTextModels to see current options)
• **temperature** (0.0-2.0): Controls randomness in the output
• Lower values (0.1-0.7) = more focused and deterministic
• Higher values (0.8-2.0) = more creative and random
• **top_p** (0.0-1.0): Controls diversity via nucleus sampling
• Lower values = more focused on likely tokens
• Higher values = considers more token possibilities
• **system**: System prompt to guide the model's behavior and personality

Customizing Text Generation

// Example options for respondText
const options = {
  model: "openai",           // Model selection
  temperature: 0.7,          // Balanced creativity
  top_p: 0.9,               // High diversity
  system: "You are a helpful assistant that explains things clearly and concisely."
};

Configuration Examples

In your MCP configuration, set defaults under env so the server uses them automatically:

{
  "mcpollinations": {
    "env": {
      "TEXT_MODEL": "openai",
      "TEXT_TEMPERATURE": "0.7",
      "TEXT_TOP_P": "0.9",
      "TEXT_SYSTEM": "You are a helpful coding assistant."
    }
  }
}

Image-to-Image Generation (NEW!)

MCPollinations now supports powerful image-to-image generation with two specialized tools:

editImage Tool

Perfect for modifying existing images:

• Remove objects: "remove the cat from this image"
• Add elements: "add a dog to this scene"
• Change backgrounds: "replace the background with mountains"
• Style modifications: "make the lighting more dramatic"

generateImageFromReference Tool

Perfect for creating variations and new styles:

• Style transfer: "make this photo look like a painting"
• Format changes: "convert this to a cartoon style"
• Creative variations: "create a futuristic version of this"
• Artistic interpretations: "make this look like a sketch"

Supported Models

• **kontext**: Specialized model optimized for image-to-image tasks
• **nanobanana**: New Google model supporting both text-to-image and image-to-image generation
• **seedream**: New ByteDance model supporting both text-to-image and image-to-image generation

Multi-reference images: editImage and generateImageFromReference accept imageUrl as a single URL or an array of URLs. The server encodes arrays as the comma-separated image parameter used by the API. Ordering matters; kontext uses only the first image, nanobanana is safe up to ~4 refs, and seedream supports up to 10.

Important: URLs only. The image-to-image tools require publicly accessible HTTP(S) URLs. Local file paths, file uploads, and base64/data URLs are not supported by this MCP server (it does not upload files). If you need to work from a local image, host it somewhere accessible (e.g., a temporary file host, object storage, or a raw link in a repo) and pass the URL.

Example Usage

// Edit an existing image
const editResult = await editImage(
  "change the background to a sunset beach",
  "https://example.com/photo.jpg",
  "nanobanana"  // or "kontext", "seedream"
);

// Generate from reference
const referenceResult = await generateImageFromReference(
  "make this into a watercolor painting",
  "https://example.com/photo.jpg",
  "seedream"  // or "kontext", "nanobanana"
);

Image Generation Details

Default Behavior

When using the generateImage tool:

• Images are saved to disk by default as PNG files
• The default save location is the current working directory where the MCP server is running
• The 'flux' model is used by default
• A random seed is generated by default for each image (ensuring variety)
• Base64-encoded image data is always returned, regardless of whether the image is saved to a file

Customizing Image Generation

// Example options for generateImage
const options = {
  // Model selection (defaults to 'flux')
  // Available models: "flux", "turbo", "kontext", "nanobanana", "seedream"
  model: "flux",

  // Image dimensions
  width: 1024,
  height: 1024,

  // Generation options
  seed: 12345,  // Specific seed for reproducibility (defaults to random)
  enhance: true,  // Enhance the prompt using an LLM before generating (defaults to true)
  safe: false,  // Content filtering (defaults to false)

  // File saving options
  saveToFile: true,  // Set to false to skip saving to disk
  outputPath: "/path/to/save/directory",  // Custom save location
  fileName: "my_custom_name",  // Without extension
  format: "png"  // png, jpeg, jpg, or webp
};

Where Images Are Saved

When using Claude or another application with the MCP server:

1. Images are saved in the current working directory of where the MCP server is running, not where Claude or the client application is installed.

2. If you start the MCP server manually from a specific directory, images will be saved there by default.

3. If Claude Desktop launches the MCP server automatically, images will be saved in Claude Desktop's working directory (typically in an application data folder).

💡 Windows Users: For reliable file saving on Windows, use absolute paths in your MCP configuration instead of relative paths (e.g., C:\Users\YourName\Pictures\MCPollinations instead of ./mcpollinations-output). Relative paths may not resolve as expected depending on the working directory context.

Finding Your Generated Images

• The response from Claude after generating an image includes the full file path where the image was saved
• You can specify a familiar location using the outputPath parameter
• Best practice: Ask Claude to save images to an easily accessible folder like your Pictures or Downloads directory

Unique Filenames

The MCP server ensures that generated images always have unique filenames and will never overwrite existing files:

1. Default filenames include:

• A sanitized version of the prompt (first 20 characters)
• A timestamp
• A random suffix

2. Custom filenames are also protected:

• If you specify a filename and a file with that name already exists, a numeric suffix will be added automatically
• For example: sunset.png, sunset_1.png, sunset_2.png, etc.

This means you can safely generate multiple images with the same prompt or filename without worrying about overwriting previous images.

Accessing Base64 Data

Even when saving to a file, the base64-encoded image data is always returned and can be used for:

• Embedding in web pages (``)
• Passing to other services or APIs
• Processing in memory without filesystem operations
• Displaying in applications that support data URIs

For Developers

If you want to use the package in your own projects:

# Install as a dependency
npm install @pinkpixel/mcpollinations

# Import in your code
import { generateImageUrl, generateImage, repsondText, respondAudio, listTextModels, listImageModels, listAudioVoices } from '@pinkpixel/mcpollinations';