A standalone Model Context Protocol server for knowledge graph operations with support for knowledge extraction, semantic search, concept analysis, and deduplication
Server has 4 tools with reasonable coverage but moderate quality issues. Tool names follow verb_noun convention (process_knowledge, search_knowledge_graph, search_concepts, get_knowledge_graph_stats). All tools have descriptions present. Input schemas are visible and mostly well-structured, though some lack complete descriptions or have design concerns. The process_knowledge tool lacks parameter descriptions beyond the source code snippet shown. Error handling strategy is basic but present. Composition is reasonable, tools are single-purpose though search_knowledge_graph has complex parameter overloading that could confuse agents.
Get statistics about the knowledge graph
Extract knowledge triples from text and store them in the knowledge graph
Search for concepts in the knowledge graph
Search the knowledge graph using fusion search
Missing output schema documentation for all search tools (search_knowledge_graph, search_concepts, get_knowledge_graph_stats). LLMs cannot plan downstream tool chains without knowing what fields to expect in results.
search_knowledge_graph weights parameter has unbounded numeric values. Missing constraints on min/max weight values. An LLM could pass weights={'entity': 10000} causing unexpected behavior.
search_knowledge_graph and search_concepts lack pagination parameters (limit, offset, cursor). If graph contains thousands of entities or concepts, results could exhaust context window. No documentation of result limits.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | B | 71 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 49 | 2024-11-05+ | v1 |
process_knowledge lacks explicit error handling strategy. What happens if extraction fails? Partial success? No recovery guidance documented.
search_knowledge_graph threshold default of 0 means all results match. This violates mxe:enforce-result-limits pattern, could return 500+ results causing context bloat.
abstraction parameter in search_concepts (high/medium/low) lacks business meaning explanation. LLM cannot determine when to use which level without understanding the hierarchy.