Loading ToolBench
MCP server enabling persistent memory for AI models through a local knowledge graph
Server has 9 tools with complete input schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (create_entities, delete_relations, search_nodes). Descriptions are present but brief (10-60 chars), below the 50-200 char baseline for LLM optimization. Parameters have type definitions and descriptions, but lack constraints (enums, ranges, patterns). No output schemas documented. Error handling is minimal, no recovery guidance or actionable error messages. Batch operations are well-designed (arrays of objects), but missing pagination for read_graph and search_nodes. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels.
Add new observations to existing entities in the knowledge graph
Create multiple new entities in the knowledge graph
Create multiple new relations between entities in the knowledge graph. Relations should be in active voice
Delete multiple entities and their associated relations from the knowledge graph
Delete specific observations from entities in the knowledge graph
Delete multiple relations from the knowledge graph
Open and retrieve specific nodes from the knowledge graph by name
Descriptions are too brief (10-60 chars vs. 50-200 char baseline). LLMs cannot determine when to select tools or understand nuanced differences between similar operations (e.g., search_nodes vs. open_nodes).
No output schemas documented. LLMs cannot plan downstream operations or extract required fields. read_graph and search_nodes return KnowledgeGraph objects but structure is not declared in tool metadata.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. Agents cannot distinguish safe reads from destructive deletes without parsing descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Read the entire knowledge graph
Search for nodes in the knowledge graph by query string
No pagination support for read_graph and search_nodes. Large knowledge graphs will return unbounded results, exhausting context windows and degrading LLM reasoning.
Error handling lacks recovery guidance. Exceptions (e.g., 'Entity with name X not found') are thrown but not caught with actionable messages. LLMs receive raw errors with no next-step guidance.