Argus MCP server has severe definition quality gaps. Tool descriptions are minimal (10-50 chars, well below the 50-200 char baseline for LLM-optimized descriptions). Most tools lack visible input schemas in the source code, only 3 of 8 tools show explicit schema definitions. Parameter descriptions are generic or missing. No output schemas documented. Error handling is absent. The server appears to expose design analysis capabilities but fails to provide the structured, self-documenting interface required for reliable agent integration.
Apply requires browser context. Use the gateway or browser extension to apply CSS changes.
Run full design + accessibility audit on a page snapshot
Compare two design score reports and return differences
Check installation health and return diagnostic information
Inspection requires browser context — return guidance
Read or export design preference profile
Score design quality of a page snapshot
Get actionable improvement suggestions based on design scoring
No input schemas visible for 5 of 8 tools (design.inspect, design.apply, design.doctor, and others). Cannot verify parameter types, constraints, or validation.
Tool descriptions are too short (10-50 chars vs. 50-200 char baseline). 'Inspection requires browser context, return guidance' and 'Check installation health and return diagnostic information' lack actionable context for LLM tool selection.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what fields does design.score return? Is there a score value, categories breakdown, recommendations?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No error handling guidance. Tools provide no recovery hints (e.g., 'If snapshot is invalid, call design.inspect first'). Agents cannot self-correct on failures.
Parameter descriptions are generic. 'PageSnapshot object containing page structure and styling data' does not explain what fields PageSnapshot must have, what format, or how to obtain one. LLMs cannot construct valid calls.