mcp_kandinsky

# MCP Kandinsky MCP Kandinsky is an MCP (Model Context Protocol) server that provides a tool for generating images using the [Kandinsky](https://www.sberbank.com/promo/kandinsky/) neural network via the official API. Currently, Kandinsky 3.1 is available via the API. ## Features - Image generation from a text description (prompt) - Style selection support (DEFAULT, KANDINSKY, UHD, ANIME) - Negative prompt support - Image size adjustment (width, height) - Saving the result to a file with overwrite control - Explicit specification of the project folder for saving (project_dir) ## Environment variables To work with MCP Kandinsky, you will need authorization data, which must be stored in environment variables. - `KANDINSKY_API_KEY` — your API key - `KANDINSKY_SECRET_KEY` — your Secret key You can obtain authorization data after registering on the [FUSION BRAIN](https://fusionbrain.ai/) website, in the API section. ### Tariffs The Kandinsky API provides a free plan with a limit of **100 requests per month**. More information about tariffs and features can be found on the [developer page](https://fusionbrain.ai/developers/). ## How to connect MCP Kandinsky to an agent (Cursor, Windsurf, etc.) Add the MCP server to the agent's config, for example, to the `.cursor/mcp.json` file or similar for your IDE: ```json "mcpServers": { "mcp-kandinsky": { "command": "uvx", "args": [ "--from", "mcp_kandinsky", "mcp-kandinsky" ], "enabled": true, "env": { "KANDINSKY_API_KEY": "ВАШ_API_KEY", "KANDINSKY_SECRET_KEY": "ВАШ_SECRET_KEY" } } } ``` ## Example of using the tool by an agent **Calling the tool from an agent (e.g., via Cursor):** ``` mcp-kandinsky/kandinsky_generate_image Arguments { "prompt": "Весенняя Москва в мае, сирень, дождь и солнце, в стиле акварельной живописи", "filename": "moskva_v_mae.png", "project_dir": "/Users/you/myproject", "width": 1024, "height": 1024, "style": "DEFAULT", "negative_prompt": "", "overwrite": false } Response Изображение успешно сгенерировано и сохранено в kandinsky/moskva_v_mae.png. ``` **Explanations:** - All parameters are passed through the agent interface. - `project_dir` is the absolute path to the project folder in which the agent is currently running. The `kandinsky` folder will be created inside it. - The response is a string with the result (success or reason for refusal). - The file is saved to the `kandinsky` folder in the project's working directory. ## Tool description **kandinsky_generate_image** | Parameter | Type | Description | |------------------|--------|-----------------------------------------------------------------| | prompt | str | Image description | | filename | str | File name for saving (e.g., cat.png) | | project_dir | str | Absolute path to the project folder | | width | int | Width (default 1024) | | height | int | Height (default 1024) | | style | str | Style (DEFAULT, KANDINSKY, UHD, ANIME; default DEFAULT) | | negative_prompt | str | Negative prompt (optional) | | overwrite | bool | Overwrite the file if it already exists (default False)| **Returns:** a result string with a message about success or the reason for refusal. ## Logic of operation - Checks for the existence of the `kandinsky` folder inside `project_dir`, creates it if necessary. - If a file with the same name already exists and `overwrite=False`, generation does not occur, and a message about the file's existence is returned. - If the file does not exist or `overwrite=True`, generation occurs, and the file is saved (or overwritten). - Returns a message about successful saving or the reason for refusal. --- API documentation: https://fusionbrain.ai/docs/doc/api-dokumentaciya/ ## License This project is distributed under the MIT license. See the LICENSE file for details.