Loading ToolBench
MCP server for Google Play Console: 170 tools for releases, Android vitals, earnings, ASO and store assets, with a confirmation gate on destructive writes
Single tool with severe definition gaps. Tool name 'get_stats_report' is appropriately verb-prefixed, but the description is cryptic and domain-specific (mentions 'collapsed rows', 'per-dimension totals', 'freshness block') without explaining WHAT the tool does or WHEN to use it from an LLM perspective. Input schema present but parameters lack actionable descriptions: 'kind' says 'Report kind (installs, ratings, earnings)' but does not explain what each kind returns or when to choose one; 'dimension' says 'Dimension to break down by (country, device, etc.)' but does not enumerate valid values or explain dependencies. No output schema documented. No error handling guidance. The description reads like internal API docs, not LLM-optimized guidance.
Shared body for the per-app bulk-report tools. Rows are collapsed to per-dimension totals: a month of country-sliced installs is a few thousand rows, and an agent needs the shape, not the file. The freshness block always travels with the numbers.
Tool description is cryptic and domain-specific. Phrases like 'Rows are collapsed to per-dimension totals' and 'freshness block always travels with the numbers' assume deep Google Play Console knowledge. LLMs cannot infer when to call this tool or what it returns.
Parameter 'kind' lacks enum constraint and actionable description. Description says '(installs, ratings, earnings)' but does not state these are the ONLY valid values, does not explain what each returns, and does not guide LLM selection.
Parameter 'dimension' lacks enum constraint and valid-value documentation. Description says 'e.g. country, device, etc.' but does not enumerate all valid dimensions or explain dependencies (e.g., which dimensions are valid for which 'kind' values).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | <=2025-11-25 | v2 |
No output schema documented. LLMs cannot plan downstream tool calls or extract relevant fields without knowing what fields the response contains, their types, and their meaning.
No error handling guidance. If 'kind' or 'dimension' are invalid, or if the month is out of range, the tool provides no recovery hints. LLMs cannot self-correct without actionable error messages.