A mcp server which extracts content and generate knowledge graph
This server exhibits poor definition quality with significant gaps in schema clarity, parameter descriptions, and output documentation. While 3 tools are registered with names and basic descriptions, most fall short of production standards. The primary tool (process_and_publish_kg) lacks complete parameter documentation and clear output schema specification. Tool names are reasonably verb-based, but descriptions are generic or domain-specific without actionable detail for LLM selection. No input validation, error guidance, or security considerations are evident in the tool definitions. The server's reliance on custom decorator patterns and incomplete schema introspection makes verification of actual tool registration difficult.
获取服务器的配置信息
A simple tool that returns a greeting message.
Extract content and generate knowledge graph from text input, supporting multiple export formats (graphml, gexf, png)
process_and_publish_kg combines two responsibilities (process + publish). Should be split into separate tools for composition.
No documented output schemas for any tool. LLMs cannot infer result structure, field types, or required downstream parameters. Critical for tool chaining.
Parameter descriptions lack actionable constraints. 'export_path' has no format guidance (absolute path? must parent exist? does tool create directory?). 'default' param in get_server_config accepts 'object' type but no schema on what that object should contain.
Descriptions lack context on WHEN to call each tool and HOW it differs from similar tools. E.g., greeting_tool vs. get_server_config both return dicts but serve very different purposes, not explained.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 44 | - | v1 |
No error handling guidance in any tool definition. What should LLM do if config key is missing? If graph export fails? If text input is invalid? No recovery hints.
process_and_publish_kg has complex internal configuration (top_k_entities, min_occur, min_edge_cooccurrence, etc.) hardcoded or loaded from defaults, but not exposed as tool parameters. LLM cannot control graph generation behavior.
Tool names and descriptions mix Chinese and English (e.g., '获取服务器的配置信息' for get_server_config). Inconsistent language makes it harder for LLMs to parse intent and reduces cross-region usability.