Loading ToolBench
MCP server for accessing Windows screenshots from WSL2
The server defines three tools with basic but incomplete documentation. Tool naming follows verb-noun convention (list_, get_), which is good. However, descriptions are terse (30-40 chars, below the 194-char baseline for A+ tools), parameter descriptions are minimal or missing, and output schemas are not documented. Input schemas are present and include type information, but lack detail on constraints and validation. Error handling is minimal, no guidance on recovery or error classification. The tools are read-only (safe) but lack the richness expected of production-grade definitions.
Get a specific screenshot file path or content
List all configured screenshot directories
List recent screenshots from Windows directories
Descriptions are under 50 characters, well below the 194-char baseline. Examples: 'List recent screenshots from Windows directories' (45 chars), 'Get a specific screenshot file path or content' (46 chars), 'List all configured screenshot directories' (40 chars).
Output schemas are not documented. LLMs cannot plan downstream calls or extract data without knowing what fields are returned. The tools return JSON objects (file paths, lists) but the response structure is not declared in the tool definition.
Parameter descriptions are sparse or missing rationale. 'limit' has a basic description, but 'pattern' lacks guidance on glob syntax rules. 'directory' says 'optional' but does not explain the behavior when omitted (searches all dirs). 'return_content' lacks explanation of base64 encoding implications (token cost, when to use).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 44 | - | v1 |
get_screenshot accepts a 'path' parameter but no validation or constraint documentation. No guidance on path format (absolute, relative, WSL vs Windows), character restrictions, or size limits. No error message template for invalid paths.
No error handling guidance. The code (visible in index.ts) uses glob and file I/O; these can fail silently or return empty results. Descriptions do not explain failure modes (permission denied, path not found, empty results) or recovery steps. LLMs have no guidance on what to try next.
list_screenshots does not document pagination or result limiting behavior. It accepts 'limit' but no 'offset' or 'cursor'. If > 20 screenshots exist, the response is truncated; LLMs cannot iterate over larger result sets. No 'total_count' or 'has_more' field documented.