EverArt

Exposes image generation capabilities through the Model Context Protocol by implementing a standardized MCP server that routes generation requests to multiple underlying AI image models (e.g., DALL-E, Stable Diffusion, Midjourney). The server translates MCP tool calls into model-specific API requests, handles authentication per model, and returns generated images through the MCP response protocol, enabling LLM clients to invoke image generation as a native tool without direct API knowledge.

Registers image generation as a discoverable MCP tool by defining a JSON schema that describes input parameters (prompt, model, size, style options) and output structure. The server exposes this schema through MCP's tools/list endpoint, allowing MCP clients to dynamically discover available image generation parameters and constraints without hardcoding knowledge of the API. This enables clients to build dynamic UIs or validate requests before sending them to the server.

Translates normalized image generation requests (generic prompt, size, style parameters) into model-specific API calls by maintaining adapter logic for each supported image generation service. When a client sends a request, the server maps generic parameters to the target model's API format (e.g., converting 'style: cinematic' to Stable Diffusion's LoRA syntax or DALL-E's style parameter), handles model-specific constraints (e.g., size restrictions), and routes the request to the appropriate API endpoint with correct authentication headers.

Implements the MCP server lifecycle by initializing the protocol transport (stdio or HTTP), registering available tools, handling incoming tool calls from MCP clients, executing image generation requests, and returning results through the MCP response protocol. The server follows MCP's request-response pattern where clients send tool calls with parameters, the server processes them asynchronously (or synchronously depending on implementation), and returns structured responses with results or errors.

Manages API credentials for multiple image generation services (e.g., OpenAI, Stability AI, Replicate) by storing them securely (environment variables or config files) and injecting them into requests to the appropriate service. The server maintains a credential registry that maps model names to their required authentication headers or API keys, ensuring that requests to each service include correct credentials without exposing them in client requests or logs.

Validates incoming image generation requests against model-specific constraints (e.g., prompt length limits, supported image sizes, valid style options) before sending them to the underlying API. The server checks parameters against a constraint registry for each model, returns detailed validation errors if constraints are violated, and may normalize parameters (e.g., rounding image dimensions to supported values) to improve request success rates.

Processes image generation responses from multiple models (which return images in different formats and structures) into a standardized format for MCP clients. The server extracts image data (URL or base64-encoded bytes), generation metadata (timestamp, model used, seed, prompt used), and error information, then formats them into a consistent MCP response structure. This enables clients to handle images uniformly regardless of which underlying model generated them.

Catches errors from image generation APIs (rate limits, authentication failures, invalid parameters, service outages) and translates them into structured MCP error responses that clients can parse and handle programmatically. The server distinguishes between client errors (invalid parameters, authentication issues) and server errors (API outages, rate limits), provides actionable error messages, and may include retry guidance or fallback suggestions.