Server has 5 tools with partial schema visibility. compress_image and generate_favicons have detailed input schemas in mcp.json, but describe_image_ai, generate_favicon_metadata_ai, and extract_image_metadata are inferred from route files without visible explicit schema registration. Descriptions are present but generic (10-50 chars for some). Parameters lack consistency in documentation, compress_image has 16 params with descriptions, but others are sparse. No output schemas documented. Error handling and recovery guidance absent. Tool composition is reasonable (single responsibility), but parameter relationships undocumented (e.g., apiKey/provider dependency in AI tools).

• Register all 5 tools explicitly in mcp.json with full input schemas and descriptions. Currently only compress_image and generate_favicons are registered; the other three are invisible to MCP clients.
• Document output schemas for each tool. E.g., compress_image should return {file_url, file_size, format, dimensions, compression_ratio}. This enables LLMs to chain calls and extract required fields.
• Move apiKey parameters to server-side environment variables or a secure vault. Accept only provider and model names as parameters. Inject credentials server-side to prevent leakage into logs.
• Add parameter relationship documentation. E.g., 'apiKey format depends on provider: Gemini uses API key string; OpenAI uses sk-* token. Both required together.' State mutual exclusivity where applicable.
• Add error handling guidance to descriptions. E.g., compress_image: 'Returns error if file exceeds 4.5 MB or format unsupported. Retry with smaller dimensions or different format.' This guides LLM recovery.