Loading ToolBench
Analyzes ad metrics and provides recommendations for optimizing ad creative performance
This MCP server has a single tool 'analyze' with critical deficiencies across naming, schema documentation, and parameter descriptions. The tool name does not follow verb_object convention (should be 'analyze_ad_metrics' or similar). The input schema is severely incomplete, it lacks type definitions for nested objects, field descriptions for list items, and validation constraints. The tool description is moderately detailed (96 chars) but does not state what the tool returns or when to use it. The implementation is a stub returning hardcoded values with no actual analytics logic, making it unsuitable for production use. The codebase shows no MCP framework integration, no server registration, no transport layer, and no protocol implementation visible.
Analyzes ad metrics and returns top performing hook and actionable recommendations
Tool name 'analyze' violates verb_noun naming convention. Should be 'analyze_ad_metrics', 'get_ad_recommendations', or similar to convey intent and resource clearly.
Input schema for 'ad_metrics' parameter lacks type definitions for nested objects. Schema shows type='list' with a description, but does not specify the structure of list items (e.g., properties for 'ad_id', 'roas', 'hook_rate' types, required fields, constraints).
Tool description does not document output schema. LLMs cannot infer that the response contains 'top_performing_hook' (string) and 'recommendations' (list of strings). LLMs need to know what fields to expect so they can plan downstream tool calls.'
No parameter-level descriptions. The 'ad_metrics' parameter has a brief description but does not explain the required structure, expected fields, validation rules, or example format.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 19 | 2026-07-28+ | v2 |
Implementation is a stub with hardcoded return values. The analyze() method ignores the 'ad_metrics' input and returns fixed recommendations. This is not a functional tool, it cannot be tested, validated, or used in production.
No error handling or validation. The tool does not validate input structure, handle empty lists, missing fields, or invalid metric values. LLMs cannot read JSON Schema pattern fields, they rely on the description text to understand constraints.'
No MCP framework integration detected. The codebase (client.py, example_usage.py) contains only a Python class with a method, no server registration, no transport layer (HTTP, SSE, STDIO), no tool capability registration, and no MCP protocol implementation. This is a standalone module, not an MCP server.