Loading ToolBench
MCP (Model Context Protocol) interface for OWL/TTL/RDF ontologies. Supports loading ontologies, SPARQL queries, adding classes/properties/individuals, and running reasoners.
The OWL MCP server provides a functional set of tools for ontology manipulation with reasonable naming conventions and mostly complete schemas. However, there are significant gaps in parameter descriptions, missing output schema documentation, and inconsistent description quality. All 10 tools are explicitly registered with input schemas, which is a strength. However, parameter descriptions are often missing or minimal, and tool descriptions lack depth regarding when to use them vs alternatives. The server uses Chinese descriptions throughout, which reduces clarity for English-speaking developers and LLMs trained primarily on English. No error handling guidance, no output schema documentation, and missing explanations of data dependencies (e.g., load_ontology must be called before other tools work).
向本体添加一个新类，可指定父类、标签和注释。修改不会自动保存，需调用 save_ontology。
向本体添加一个新个体，并指定所属类。修改不会自动保存，需调用 save_ontology。
描述一个类：父类、子类、等价类、限制条件、所有个体。
描述一个个体：所属类、属性值（数据属性和对象属性）。
获取当前加载本体的基本信息：IRI、类数量、个体数量、属性数量等。
列出本体中所有类（OWL Class）。可选按名称前缀过滤。
列出本体中所有个体（OWL Individual）。可选按类名过滤。
All tool and parameter descriptions are in Chinese, reducing accessibility and clarity for English-speaking developers and most LLM training data
get_ontology_info has empty properties object in inputSchema, no parameters documented, but also no explanation of what 'basic info' includes (IRI format, count types, etc.)
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract needed fields without knowing response structure. Example: describe_class returns unstructured text; should document fields like parent_classes, subclasses, restrictions, individuals.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
列出本体中所有对象属性（ObjectProperty）和数据属性（DataProperty）。
加载一个 OWL/TTL/RDF 本体文件（本地路径或 IRI）。后续所有操作都在这个本体上进行。
对当前本体执行 SPARQL SELECT 查询，返回结果列表。
No error handling guidance. Tools can fail (ontology not loaded, class not found, SPARQL syntax error, write failures) but provide no recovery hints. Example: 'Class X not found' should suggest 'Try list_classes() to see available classes'.
Missing critical dependency documentation: all tools except load_ontology require an ontology to be loaded first (_check_loaded() throws RuntimeError). This is not documented in any tool description, forcing agents to discover this through trial and error.
Parameter descriptions are minimal or absent. Example: sparql_query 'query' parameter has description 'Complete SPARQL 1.1 SELECT query statement' but no guidance on SELECT vs CONSTRUCT, result format expectations, or common errors.
add_class and add_individual have incomplete parameter descriptions. parent_class defaults to 'Thing' but this is not explained, agents don't know if this is owl:Thing or a custom class. add_class notes say 'changes won't auto-save' but don't mention save_ontology tool exists or how to call it.
No tool for save_ontology, delete_class, delete_individual, or update operations. After add_class/add_individual indicate changes won't save, there is no documented path to persist them. Code shows _sync_graph_from_onto() is called but save_ontology tool is incomplete (cut off in source).
list_* tools accept 'limit' parameter defaulting to 100 but don't document what happens when results exceed the limit or offer pagination. No mention of offset, cursor, or next_page guidance.