Loading ToolBench
AI-powered financial tools and connectors with Neon DB - provides MCP tools for account management, transaction queries, analytics, budgeting, forecasting, and alerts
Server has 13 tools with complete input schemas and descriptions. Naming follows verb_noun convention (list_, get_, create_, search_). However, descriptions are generic and lack LLM-optimization guidance. Parameter descriptions exist but are minimal (avg ~40 chars vs baseline 72). No output schemas documented. Error handling and recovery guidance absent. Missing tool annotations (readOnlyHint/destructiveHint). Tools are well-composed (single responsibility) but lack chaining context (e.g., create_budget doesn't indicate what fields are returned for downstream use).
Get status of all active budgets with spending progress
Predict future cash flow based on historical patterns
Create a new financial alert
Create a new spending budget
Get full details of a specific transaction
Compare income against expenses for a period
List all connected bank accounts for the user
No output schemas documented. LLMs cannot infer what fields are returned, forcing them to guess downstream tool parameters. E.g., create_budget returns what? (budget_id, name, status?). This breaks tool chaining.
Descriptions lack LLM-optimization guidance. Generic descriptions like 'Get spending breakdown by category for a time period' don't explain WHEN to use this vs spending_trends, or what the output structure is. Baseline: 194 chars; these average ~60 chars.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2025-06-18+ | v2 |
List all configured alerts
Query transactions with filters like date range, category, amount, merchant
Get a complete financial summary for a month
Full-text search across transaction descriptions, merchants, and notes
Get spending breakdown by category for a time period
Get spending trends over multiple periods
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe read-only tools from destructive ones. create_budget and create_alert are WRITE operations but lack destructiveHint annotation.
Parameter descriptions are minimal. E.g., 'alertAt' in create_budget says 'Alert when this % is reached (e.g., 80)' but doesn't clarify: is this 0-100 or 0-1? Is it required? What happens if omitted? Baseline: 72 chars; these average ~40 chars.
No error handling or recovery guidance. If list_transactions fails (e.g., invalid date format), the response likely returns a raw error. LLMs need actionable guidance: 'Invalid startDate format. Use YYYY-MM-DD (e.g., 2024-01-15).'