Critical failures across all dimensions. Description is generic (58 chars, below 20-char floor for scoring). No parameter documentation exists. Tool performs a complex medical imaging operation but provides zero guidance on prerequisites, output structure, or error recovery. The implementation shows sophisticated DICOM processing (rescaling, windowing, PNG conversion) but this is completely hidden from the LLM, the tool interface is opaque and unusable for agentic composition.

• Add input parameters: series_id (string, required), slice_index (integer, optional, default=0), window_center (float, optional), window_width (float, optional). Each parameter must have a description explaining its purpose and valid range.
• Expand tool description to 100-150 chars: 'Display DICOM medical images from a series. Accepts optional windowing parameters (window_center, window_width) to adjust contrast. Returns PNG image as base64 and extracted metadata (patient name, series description, slice location, dimensions). Call this after identifying a series ID via search or list tools.'
• Document output schema: { image_base64: string, metadata: { filename, width, height, patient_name, series_description, slice_location, instance_number }, slice_count: integer, current_slice: integer }
• Add error handling: return structured errors with recovery hints. E.g., 'Series not found. Available series: [list]. Call list_dicom_series() to discover available data.' Distinguish between missing files (user-fixable) and parse errors (fatal).
• Implement pagination/filtering: add series_filter (string, optional) and limit (integer, 1-100, default=20) to support discovery without loading all files. Return total_series_count and next_cursor for large datasets.

Inferred effective spec: 2026-07-28+.

• Current-spec patterns present: stateless request handling, Streamable HTTP