Server has 7 tools with basic schemas and descriptions, but significant gaps limit production readiness. Image tools (create_thumbnail, detect_faces, blur) have detailed docstrings (150-250 chars) and complete parameter schemas with types and descriptions. Math tools (add, subtract, multiply, divide) have minimal descriptions (10-30 chars, below 20-char baseline) and lack parameter context. No tool declares output schemas, error handling is generic string returns without recovery guidance, and no tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles. Parameter descriptions are present but lack constraints (e.g., blur coordinates should specify 0.0-1.0 range explicitly in schema, not just docstring). STDIO transport caps protocol score at 50.

• Expand math tool descriptions to 50-100 chars each. E.g., 'Adds two integers and returns their sum. Use for arithmetic operations. Returns int. Raises ValueError if result overflows.' This gives LLM context for selection and error handling.
• Document output schemas for all tools. E.g., create_thumbnail returns {type: 'object', properties: {path: {type: 'string'}, success: {type: 'boolean'}}}. Enables LLM to chain calls and extract data.
• Add recovery guidance to error messages. E.g., 'Error: Image file not found. Check the file path exists and is readable. Try list_files() to see available images.' This guides agent retry logic.
• Add tool annotations via FastMCP metadata. Mark create_thumbnail, blur as destructiveHint: true; detect_faces, math tools as readOnlyHint: true. Enables agents to reason about side effects.
• Formalize parameter constraints in JSON Schema. Add minimum/maximum for blur coordinates, add enum for file formats, add pattern for file paths. Remove example values from descriptions.