A Model Context Protocol server for image editing
5 image editing tools with solid naming and basic descriptions, but significant gaps in parameter documentation, output schema clarity, and error handling guidance. All tools follow verb_noun convention (adjust_brightness, crop_image, etc.), which is strong. Descriptions are present but brevity leaves LLM intent unclear in edge cases. Parameter schemas exist with type definitions and constraints (e.g., quality 1-100, crop coordinates with minimum bounds), but lack rich descriptions for several fields. No documented output schemas or error recovery guidance. Tool names are action-clear but descriptions could be 2-3x longer to optimize LLM understanding. No security scoping, no error taxonomy, no confirmation patterns for destructive operations (all are WRITE operations).
Adjust the brightness of an image (supports PDF input as image)
Compress the image with specified quality (supports PDF input as image)
Convert an image (or PDF page) to a PDF file
Crop the image to a specified rectangle (supports PDF input as image)
Trim whitespace from the image edges (supports PDF input and PDF output)
Sparse parameter descriptions. Parameters like 'factor' (brightness), 'quality' (compress), and 'keepAspectRatio' (optimize) have minimal context. 'factor': '1.5 for 50% brighter' is good, but 'quality': 'Compression quality (1-100)' lacks guidance on what quality 90 vs 50 means in practice or recommended defaults for typical images.
No documented output schemas. Tools return file writes to disk but client has no structured response schema. What does compress_image return on success? A confirmation string? File metadata? This forces LLMs to infer behavior. Pattern requires explicit output documentation.
No error recovery guidance. No tool description states what to do on failure (e.g., 'File not found', 'Invalid crop bounds exceed image dimensions', 'PDF conversion failed'). Error responses should tell LLMs: is this retryable? User-fixable? Fatal? Required by error-classification pattern.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 49 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
No confirmation or dry-run for destructive operations. All 5 tools are WRITE operations that overwrite/create files. No tool offers a 'dry_run' mode or 'confirm_before_execute' pattern. Agents could accidentally overwrite user files. Pattern: confirmation-request.
Tool descriptions are under 100 chars (typical baseline 194). 'Adjust the brightness of an image...' (59 chars) is terse. LLM cannot easily distinguish when to use adjust_brightness vs optimize_image. Descriptions should explain the intent and typical use cases in 150-250 chars.
'sourceFilename' and 'outputFilename' parameters are generic string types with minimal context. LLMs may pass URLs, relative paths, or invalid formats. Descriptions should specify: absolute vs relative, required extension format, whether paths are validated, and what happens if the file doesn't exist.
crop_image exposes low-level pixel coordinates (left, top, width, height) with only 'minimum' bounds but no maximum or guidance. LLM might pass coordinates exceeding image dimensions. Description should state: 'Must not exceed image bounds; if so, operation fails with boundary error.'
No scope or permission declarations. Tools don't state what file system permissions they require or what security boundaries they respect. Required by scope-declaration pattern for compliance and least-privilege configuration.