Loading ToolBench
MCP-native demand generation server. Expose LinkedIn signal detection, ICP scoring, contact enrichment, and CRM routing as typed MCP tools.
Four tools with complete input schemas and descriptions, but significant gaps in parameter documentation and output schema specification. All tools have descriptions (10-100 chars), but three lack parameter-level descriptions entirely. No output schemas documented. Error handling is minimal, tools return mock data without guidance on failure modes or recovery. Naming follows verb_noun convention (analyze_, update_, draft_, get_) but lacks specificity about what data is extracted or modified. No tool annotations (readOnlyHint, destructiveHint) despite clear risk profiles.
Ingest a sales call recording and extract sentiment, objections, talk ratios, and stage inference.
Generate a post-call follow-up email from conversation content.
Get industry benchmark data for win rate, show rate, and cycle length.
Move a deal to a new stage in the CRM pipeline.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. Tools return mock JSON strings without formal response type definitions.
Parameter descriptions missing for all non-trivial parameters. 'stage' enum in update_pipeline lacks description; 'tone' enum in draft_followup lacks description; 'segment' in get_benchmarks lacks description. LLMs cannot infer when to use each value.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). update_pipeline is clearly destructive (modifies CRM state) but lacks destructiveHint. Agents cannot distinguish safe from risky operations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Error handling is absent. Tools return mock success responses without documenting failure modes, recovery paths, or actionable error messages. No guidance for LLM on what to do if call_id or deal_id is invalid.
Tool descriptions are generic and lack context. 'Ingest a sales call recording...' does not explain when to call this vs other tools, what prerequisites exist, or what the output enables. Descriptions should be 50-200 chars and answer WHAT, WHEN, and WHY.