Loading ToolBench
A collection of MCP-compatible skills/tools for marketing research, analytics, automation, SEO, email, and growth tasks
Scoring was not performed
Output schemas not documented. None of the 23 tools document what they return. LLMs cannot plan multi-step workflows or extract required fields without knowing the response structure.
Tool registration not visible. Only ab-test-stats/main.py is shown in source code. The actual MCP tool registration (tool names, descriptions, input schemas) for 22 of 23 tools is not provided, making verification impossible.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | F | 28 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 43 | - | v1 |
Descriptions are sparse and lack LLM-optimization. Most descriptions are 20-40 characters, well below the 194-char baseline. E.g., 'Analyze sentiment of a single text' (35 chars) lacks context on WHEN to use it, what models do, when to prefer batch vs single, or what the output contains.
Parameter descriptions are missing or vague. E.g., 'retention' has 'periods' param with enum ['week', 'month', 'quarter'] but no description explaining what 'periods' controls. 'repurpose' has 'style' enum without explaining how styles affect output.
No error handling or recovery guidance. Tools provide no error messages, validation feedback, or suggestions for LLM recovery. If 'retention' fails on a malformed CSV, what should the LLM do? If 'analyze (sentiment)' hits a model timeout, can it retry?
Tool naming collision. Tools 'analyze', 'batch', 'visualize', and 'report' appear multiple times across different skills (funnel vs cohort vs sentiment). LLMs will struggle to disambiguate, e.g., is 'analyze' for funnel or sentiment?
No pagination or result limiting documented. Tools like 'analyze (funnel)' may return large lists (all drop-off stages). The source shows no limit param, no pagination guidance, and no note on max results. Return only the fields the agent needs.'
Type consistency issues. 'confidence' is float with default 0.95 in significance tools, but no validation that 0 < confidence < 1. 'periods' enum in retention is lowercase ['week', 'month', 'quarter'], must be documented as exact strings. When a parameter has a regex pattern or character restriction, state it in the description.'