Single tool 'fetch' has a clear action verb, well-structured schema with all parameters typed and described, and a solid description (127 chars). Input schema is explicit with proper JSON schema types (string, boolean). Error handling is present (panic recovery, error wrapping). However, output schema is not formally documented, the tool returns markdown/text content but the response structure is implicit rather than explicit. Parameter descriptions are good but could be more prescriptive about constraints (e.g., timeout format, return_format enum values). No tool annotations (readOnlyHint, etc.) despite being a read-only operation.

• Add formal output schema documentation: declare that the tool returns a single TextContent field with markdown or plaintext string. Example: 'Returns: {content: string (markdown or plaintext)}'. This enables agents to extract and chain results.
• Convert return_format to an enum constraint in the schema: enum=['markdown', 'text']. Remove the example from the description and rely on the enum for clarity.
• Add a regex pattern or format constraint to the timeout parameter: pattern='^\d+[smh]$' (e.g., '30s', '5m', '1h'). Include this in the schema, not just the description.
• Add readOnlyHint=true to the tool definition to signal that fetch is idempotent and safe to retry. This helps agents understand they can call it multiple times without side effects.
• Enhance error messages with recovery guidance. For example: 'Invalid timeout format: got "30 seconds". Use Go duration format: 30s, 5m, 1h, etc.' This enables LLMs to self-correct in one retry.

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

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