Loading ToolBench
Streamable HTTP transport MCP server for querying Deeplore knowledge collections. Returns grounded answers from lecture notes and transcripts using RAG (Retrieval-Augmented Generation).
Two tools with basic schemas and descriptions. 'ask' has a dynamic enum-like description but no formal enum constraint in schema, forcing LLMs to guess valid collection names. 'list_projects' lacks parameter descriptions (empty properties object). Both tools return unstructured TextContent with no documented output schema. Error handling is present but minimal, errors are returned as plain text rather than structured recovery guidance. No tool annotations (readOnlyHint, idempotentHint). Naming is clear (verb_noun pattern), but parameter documentation is sparse.
Query a Deeplore knowledge collection. Returns a grounded answer from lecture notes and transcripts. Available collections: {dynamic list}
List all available Deeplore knowledge collections.
No enum constraint for 'collection' parameter. Description lists available collections dynamically, but schema does not enforce them as an enum. LLMs may hallucinate invalid collection names.
'list_projects' has empty inputSchema properties and no parameter descriptions. Even though it takes no parameters, the tool description should explain what it returns and when to call it (discovery pattern).
No documented output schema. Both tools return TextContent with unstructured text. LLMs cannot plan downstream calls or extract structured data. Should document return format (e.g., list of collection names, answer with sources).
Error responses are plain text strings without recovery guidance. 'Unknown collection' error should suggest available collections or recommend calling list_projects first.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
No tool annotations. Both tools are read-only and idempotent, but lack readOnlyHint and idempotentHint annotations in schema. Agents cannot infer safety properties.