Single tool 'hello_world' with basic definition. Tool name follows verb pattern (positive). Description is short but present (15 chars, below baseline 194). Input schema visible with type and description for 'name' parameter. However, output schema is not documented, LLMs cannot know what structure to expect from the response. Tool is READ_ONLY which is appropriate but not explicitly annotated in schema. The codebase shows HTTP/SSE transport via mcp-go framework, but core server implementation is minimal, a proof-of-concept rather than production-grade.

• Expand tool description from 15 to 50-100 characters. Add context: 'Greets a person by name and returns a personalized greeting. Use this for simple text generation tasks. Input: person's name (string). Output: greeting text.'
• Document output schema: specify that CallToolResult contains a 'text' field with the greeting string.
• Add tool annotations to schema: include 'readOnlyHint': true since tool is READ_ONLY and has no side effects.
• Improve error message: return 'Invalid input: name must be a string. Received: {actual_type}. Example: "Alice"' to guide the LLM on correction.
• Add parameter constraint in description: 'Name should be 1-100 characters, no special symbols.' This prevents invalid inputs before validation.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry