MCPollinations Multimodal MCP Server

📜 Changelog All notable changes to the MCPollinations will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [1.2.2] - `2025-09-26` ### Added - **New Image Models**: Added support for two new image generation models - `nanobanana`: Google's new model supporting both text-to-image and image-to-image generation - `seedream`: ByteDance's new model supporting both text-to-image and image-to-image generation - Both models work identically to `kontext` for image-to-image operations (editImage, generateImageFromReference) - Full support for all existing parameters: dimensions, seed, enhancement, safety filtering, file saving - **Enhanced Documentation**: Updated README.md and OVERVIEW.md to include new models - Added examples showing usage with new models - Updated supported models lists in all documentation - **Updated Configuration**: Enhanced MCP configuration generator to include new models in available options - **Test Suite**: Added comprehensive test files for MCP functionality - `test-mcp-client.js`: Basic MCP protocol connectivity test - `test-list-tools.js`: Tool enumeration and schema validation test - `test-image-save.js`: Image generation and file saving functionality test - `test-new-models.js`: Validation test for new image models - `test-image-to-image.js`: Image-to-image functionality test with new models ### Changed - **Schema Updates**: Updated tool schemas for `editImage` and `generateImageFromReference` to include new models in descriptions - **Dynamic Model Support**: All new models are automatically available through the existing `listImageModels` API endpoint - **Backward Compatibility**: All existing functionality remains unchanged - new models are additional options ### Technical Details - New models use identical API patterns to existing models - Support for up to 4 input images (following Pollinations API patterns) - All existing parameters (width, height, seed, enhance, safe, etc.) work with new models - File saving, unique naming, and base64 encoding work identically across all models ## [1.2.1] - `2025-09-06` ### ⚠️ BREAKING CHANGES - **Removed `gptimage` model**: The `gptimage` model is no longer supported by the Pollinations API - All image-to-image operations (editImage, generateImageFromReference) now use `kontext` as the default model - Configuration files and examples have been updated to reflect available models - **Removed transparent background support**: Transparent background functionality has been removed - Removed `transparent` parameter from all image generation tools - Updated tool schemas to remove transparent parameter - Removed transparent background configuration options - Documentation updated to remove all references to this feature ### Changed - **Image-to-Image Model Defaults**: Updated default models for image-to-image operations - `editImage` tool: Default model changed from 'gptimage' to 'kontext' - `generateImageFromReference` tool: Default model changed from 'gptimage' to 'kontext' - Text-to-image operations continue to use 'flux' as default - **Configuration Updates**: Updated all configuration files and generators - `example-mcp.json`: Removed transparent parameter - `generate-mcp-config.js`: Removed transparent configuration prompts and gptimage references - Updated available model lists in configuration generator - **Documentation Updates**: Comprehensive documentation cleanup - `README.md`: Removed all references to gptimage model and transparent backgrounds - Updated code examples to use supported models only - Removed deprecated feature sections ### Removed - `transparent` parameter from all image generation functions and schemas - `gptimage` model references from documentation and configuration files - Transparent background configuration prompts - Deprecated feature documentation sections ### Migration Guide - **For users with existing configurations**: Remove any `transparent` parameters from your MCP configuration files - **For image-to-image operations**: Replace any `gptimage` model references with `kontext` - **For transparent backgrounds**: This functionality is no longer available - consider using image editing tools as a post-processing step ## [1.2.0] - `2025-07-25` ### Added - **Transparent Background Support**: Added `transparent` parameter for generating images with transparent backgrounds - Works with `gptimage` model only - Perfect for logos, icons, and graphics that need transparent backgrounds - Added to both `generateImageUrl` and `generateImage` tools - Configurable in MCP configuration with default settings - **Image-to-Image Generation**: Two new powerful tools for working with existing images - `editImage` tool: Edit or modify existing images based on text prompts (e.g., "remove the cat and add a dog", "change background to mountains") - `generateImageFromReference` tool: Generate new images using existing images as reference (e.g., "make this into a cartoon", "create a painting version") - Both tools support `gptimage` and `kontext` models - Full parameter support including transparent backgrounds, custom dimensions, and file saving - **Enhanced Configuration**: Updated configuration generator and example files - Added transparent parameter configuration prompts - Updated tool lists to include new image-to-image tools - Enhanced model guidance and descriptions ### Changed - Updated tool schemas to include new transparent parameter - Enhanced image generation capabilities with more flexible model options - Improved tool descriptions for better AI assistant understanding ### Fixed - Ensured all new tools follow consistent parameter patterns and error handling ## [1.1.3] - `2025-07-25` ### Added - **Enhanced Text Generation**: Added configurable parameters for text generation - `temperature` parameter (0.0-2.0) for controlling randomness in output - `top_p` parameter (0.0-1.0) for controlling diversity via nucleus sampling - `system` parameter for providing system prompts to guide model behavior - Configuration generator now includes prompts for text generation parameters - **User Configuration Priority**: Added documentation and tool descriptions emphasizing user-configured settings are used as defaults - **Improved Model Guidance**: Updated tool schemas to reference listTextModels and listImageModels for current model lists - **Text Generation Privacy**: Added hardcoded `private=true` parameter to text generation requests ### Changed - Updated configuration generator with more accurate model information and tier-specific messaging - Enhanced tool descriptions to clarify user config priority and override behavior - Improved path guidance for Windows users in configuration prompts ### Fixed - Added missing `private=true` parameter to text generation API requests ## [1.1.2] - `2025-07-25` ### Added - **Optional Authentication Support**: Added support for optional `token` and `referrer` parameters to access enhanced Pollinations API features - Environment variable support: `POLLINATIONS_TOKEN` and `POLLINATIONS_REFERRER` - MCP configuration file support with new `auth` section - Authorization header (`Bearer token`) and Referer header support in all API requests - Backward compatible - free tier continues to work without authentication - Enhanced MCP configuration generator with authentication prompts - Updated example-mcp.json to include auth section template - Comprehensive documentation updates for authentication setup - **Windows Path Guidance**: Added documentation for Windows users to use absolute paths for reliable file saving ### Changed - All service functions now accept optional `authConfig` parameter - MCP server now reads and passes authentication configuration to services - Configuration generator includes new authentication configuration section ### Fixed - Improved file path handling documentation for Windows compatibility ## [1.0.7] - `2025-04-08` ### Added - Project analysis and `OVERVIEW.md` update by Pink Pixel on `2025-04-08`. - Added Usage and Key Features sections to OVERVIEW.md. - Comprehensive codebase analysis. - Added ability to save generated images to a customizable file path as PNG (or other formats). - New options for `generateImage` tool: `saveToFile`, `outputPath`, `fileName`, and `format`. - Added hardcoded parameter `nologo=true` for all image generation. - Added customizable parameter `safe` for content filtering (defaults to false). - Set default seed to random for image generation to ensure variety. - Added customizable parameter `enhance` for image quality enhancement. - Set 'flux' as the default model for image generation. - Changed default behavior to save images to file automatically (PNG format). - Added comprehensive documentation in README.md about image saving behavior, locations, and accessing base64 data. - Implemented unique filename generation to prevent overwriting existing images. - Added automatic numeric suffixes for duplicate filenames. - Renamed `generateText` tool to `respondText` for clarity and consistency. - Removed `sayText` tool as it's not supported by the API. - Removed generic `listModels` tool in favor of specific `listImageModels` and `listTextModels` tools for clarity. - Ensured `model` parameter is properly documented as customizable for text generation. - Improved `listAudioVoices` tool to return the complete list of available voices. - Enhanced documentation for the `voice` parameter in `respondAudio` tool. - Fixed tool registration to ensure all tools are properly displayed in the MCP protocol. - Added test script to verify tool registration. - Standardized package name references throughout the codebase. - Replaced Claude-specific installation script with a comprehensive MCP configuration generator that supports customizing: - Output and temporary directories (using relative paths for portability) - Default parameters for image, text, and audio generation - Tool restrictions and permissions - Fixed server and package names to ensure compatibility - Removed nologo parameter from configuration as it's hardcoded to true - Added lists of available models for image and text generation to help users make informed choices - Changed the command to start the server from "model-context-protocol" to "mcpollinations" for better branding consistency - Removed `listPrompts` and `listResources` tools as they're not currently implemented with an API - Removed resourceService.js and resourceSchema.js files - Updated tool schemas to expose only the parameters we want to customize in the MCP client - Added MIT LICENSE file - Added comprehensive .gitignore file - Updated documentation to reflect the new configuration generator. ## [1.0.6] - `2025-04-01` ### Added - Compatibility with Node.js versions 14.0.0 and later. - AbortController polyfill for Node.js versions below 16.0.0. - Troubleshooting guide in README. - Enhanced documentation with system requirements and installation options. ### Fixed - "AbortController is not defined" error. - Improved error handling and reporting. ## [1.0.5] - `2025-04-01` ### Added - Initial public release. ## [1.3.0] - `2025-09-26` ### Added - Multi-reference image support for new models: - `editImage` and `generateImageFromReference` now accept `imageUrl` as a string or an array of URLs. Arrays are encoded as the API’s comma-separated `image` parameter. - Environment-driven defaults used when tool calls omit arguments: - Image: `IMAGE_MODEL`, `IMAGE_WIDTH`, `IMAGE_HEIGHT`, `IMAGE_ENHANCE`, `IMAGE_SAFE` - Text: `TEXT_MODEL`, `TEXT_TEMPERATURE`, `TEXT_TOP_P`, `TEXT_SYSTEM` - Audio: `AUDIO_VOICE` - Files: `OUTPUT_DIR` ### Changed - Authentication loading recognizes both `token`/`referrer` (lowercase) and `POLLINATIONS_TOKEN`/`POLLINATIONS_REFERRER` (uppercase) from env, matching MCP client configs that pass simple keys. - Configuration generator now outputs an `env`-only MCP config. Removed legacy `resources`, `disabled`, and `alwaysAllow` blocks to align with typical MCP client expectations. - Example configuration and docs updated to use `env` for auth and defaults. ### Fixed - Ensured server respects user defaults from env across all tools, so user’s configuration is always honored unless explicitly overridden in a request. ### Notes - Model behavior details for `kontext`, `nanobanana`, and `seedream` are documented in `new_models_integration.md`. `kontext` uses only the first reference; `nanobanana` is safe up to ~4 refs; `seedream` up to 10. ## [1.3.1] - `2025-09-26` ### Fixed - Image-to-image query construction for kontext, nanobanana, and seedream: - Now sends multiple reference images as repeated `image` parameters instead of a single comma-encoded string. This matches the API examples and fixes 400/403 errors some users saw when passing multiple references. - Continues to accept a single URL or a comma-separated string; comma strings are split and sent as repeated params.