Loading ToolBench

Reading the current corpus…

snapgrab MCP server · C · ToolBench

← Browse

snapgrab

github.com/ArkNill/snapgrab

URL to screenshot with metadata. Python MCP server included. Claude Vision optimized.

2026-07-28+STDIO3 tools

69/100

Definition Quality77

Protocol Readiness50

SupportabilityNot measured-

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Three tools with complete input schemas and descriptions. Naming follows verb_noun pattern (capture_*, extract_*). Descriptions are adequate (100-200 chars) but lack WHEN-to-use guidance and dependency hints. Parameters are well-typed with defaults, but lack enum constraints for viewport/format/wait values. Output is JSON-serialized but schema not formally documented. No error handling guidance or recovery paths. Missing tool annotations (readOnlyHint, idempotentHint). Composition is sound, each tool has one responsibility.

Tools (3)

capture_comparisonread onlysource verified75/100

Capture screenshots of the same URL in multiple viewports for comparison. Useful for responsive design review — compare desktop vs mobile layouts.

capture_screenshotread onlysource verified80/100

Capture a screenshot of a web page and return the file path. The saved image can be viewed with the Read tool. Also returns page metadata (title, OG tags, favicon) and Claude Vision token estimate.

extract_page_metadataread onlysource verified77/100

Extract page metadata without taking a screenshot. Returns title, description, Open Graph tags, favicon URL, and HTTP status. Faster than capture_screenshot when you only need metadata.

Issues (5)

HIGHconstrained-input ↗

Viewport, format, and wait parameters accept free-form strings instead of enums. LLMs may hallucinate invalid values like 'ultrawide' or 'webp'.

capture_screenshotcapture_comparison

HIGHresponse-shaper ↗

Output schemas not formally documented. Tools return JSON strings, but LLMs cannot parse the structure without explicit schema declaration.

capture_screenshotcapture_comparisonextract_page_metadata

HIGHrecovery-guide ↗

No error handling guidance. If a URL is unreachable or times out, tools return raw exceptions. LLMs have no recovery path.

capture_screenshotcapture_comparisonextract_page_metadata

MEDIUMtool-description ↗

Descriptions lack WHEN-to-use context. 'Capture a screenshot' is WHAT; missing is WHEN to call this vs extract_page_metadata or capture_comparison.

capture_screenshotcapture_comparison

Recommendations

Convert viewport, format, and wait parameters to enums: viewport: ['desktop', 'tablet', 'mobile'], format: ['png', 'jpeg', 'pdf'], wait: ['networkidle', 'domcontentloaded', 'load'].
Document output schemas explicitly. Define the structure of the JSON response (path, metadata fields, vision_tokens, etc.) so LLMs know what to expect.
Add error handling: wrap capture() and extract_metadata() calls in try-except, return structured errors like {"error": "timeout", "message": "URL did not respond within 30s", "retry": true}.
Enhance descriptions with WHEN-to-use guidance: 'Use capture_screenshot for full-page visual review. Use extract_page_metadata if you only need title/description/OG tags (faster, no image).'
Add tool annotations via FastMCP: @mcp.tool(annotations={"readOnly": true, "idempotent": true}) to declare safety properties.

Score history

Overall score trend

First recorded score · v2 rubric

69/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	69	2026-07-28+	v2