Loading ToolBench
Gridproof your UI — automated spacing & grid QA in the agent loop (MCP server).
Gridproof registers 4 tools (gp_audit, gp_check_element, gp_get_config, gp_report) with clear verb-noun naming and reasonable descriptions. All tools have documented input schemas with proper types and descriptions. However, output schemas are not formally documented in the visible source code, and descriptions lack detail on return structure, making it harder for LLMs to plan downstream operations. Error handling and recovery guidance are minimal. Tool descriptions are adequate but could be more LLM-optimized (10 - 200 char target). Parameters mostly have types and descriptions, but schema documentation for return values is absent.
Run an audit on a URL and return raw spacing/grid findings. Loads config, renders the page, collects geometry, and runs enabled rules.
Check a single element's spacing/sizing against grid rules. Returns violations for a specific selector.
Load and return the active Gridproof configuration (merged from defaults and gridproof.config.json if found).
Run an audit and write a styled HTML report file. Returns the report path and summary.
Output schemas not documented. Tool descriptions do not explain what fields are returned (e.g., gp_audit returns 'violations' array structure, gp_report returns 'report path and summary' but no field names). LLMs cannot plan multi-tool chains without knowing what data they'll receive.
Error handling and recovery guidance missing. Tools have no documented error conditions or actionable error messages. No guidance on what LLM should do if audit fails (network error, invalid CSS selector, Playwright timeout, config parse error).
Parameter descriptions lack detail on format and constraints. 'viewport' parameter documented as 'Viewport dimensions' but no minimum/maximum values specified. 'rules' array described as 'Optional list of rule IDs' but no examples of valid rule IDs or constraints on array length provided. 'selector' parameter lacks CSS selector validation rules or error cases.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
gp_get_config description too generic ('Load and return the active Gridproof configuration'). LLM cannot infer when to call this vs using audit tools directly. No guidance on what config controls or how changes affect audit behavior.
gp_report tool marked READ_ONLY but actually has WRITE risk (creates/writes file to disk). Risk classification is inverted, potentially confusing agents about side effects.