Loading ToolBench
Neo4j MCP server has critical definition gaps. Tool descriptions are minimal (10-20 chars, well below 50-200 char baseline for LLM optimization). Input schemas present but lack depth: parameters have types and basic descriptions, but no enums, constraints, ranges, or format specifications. No output schemas documented. Error handling is generic (try/except with logging) but provides no recovery guidance to LLMs. Tool names are verb-noun compliant (execute_query, create_node, create_relationship), but descriptions fail to explain WHEN to use each tool, what prerequisites exist, or what the return structure contains. No parameter validation rules stated. No idempotency guarantees. Security concerns: credentials passed via environment (good), but no audit logging or permission gates visible.
创建节点
创建关系
执行Cypher查询
Descriptions critically short (10-20 chars). 'Execute Cypher query', 'Create node', 'Create relationship' lack context on WHEN to use, prerequisites, or return structure. Baseline for LLM-optimized descriptions is 50-200 chars.
No output schemas documented. LLMs cannot infer what fields execute_query returns (list of dicts? field names?), what create_node returns (node ID? full node object?), or what create_relationship returns. Breaks downstream tool chaining.
No parameter constraints or validation rules stated. 'query' param accepts any string (risk of injection). 'label' and 'rel_type' have no enum or format constraints. 'properties' is untyped object with no schema. LLMs cannot validate inputs before calling.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 36 | <=2025-11-25 | v2 |
Error handling provides no recovery guidance. Exceptions logged but LLM receives raw error messages ('Neo4j query error: ...') with no actionable next steps. No error classification (retryable vs fatal).
No idempotency guarantees or confirmation steps for destructive operations. create_node and create_relationship modify state but lack dry-run, confirmation, or idempotency hints. Agents cannot safely retry on ambiguous failures.