Loading ToolBench
Academic paper knowledge-base MCP tool with ingest, retrieval and chat over your document corpus
Scholar-RAG has 11 tools with complete input schemas and descriptions. Naming follows verb_noun convention (search_*, list_*, get_*, create_*, delete_*). Descriptions are present and adequate (avg ~100 chars), but lack LLM-optimized guidance on WHEN to use each tool vs. alternatives. Parameters are typed and described, but lack constraints (enums, ranges, patterns). Output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Error handling is absent from tool definitions. Destructive operations (remove_document, delete_kb) lack confirmation patterns. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels.
Asynchronously ingest a single PDF into an existing kb; returns a job_id.
Asynchronously create a kb from all PDFs in a folder; returns a job_id.
Two-phase kb deletion: preview and confirm token, else full deletion.
Overview of a document: metadata, abstract, section outline and total character count.
Paginated reading of the full text or a single section of a document.
Query the status, progress and result of a background job.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream tool chaining. E.g., search_chunks returns embedding/rerank scores but no schema tells the LLM what fields to expect.
Destructive operations (remove_document, delete_kb) lack confirmation or dry-run patterns. delete_kb has a confirm_token parameter but no description of how to obtain it or what happens without it. Agents may accidentally delete knowledge bases.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tools are marked with Risk levels in metadata but not exposed to LLM via MCP tool annotations. LLM cannot reason about safety without explicit hints.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | <=2025-11-25 | v2 |
Paginated browse of documents in a kb.
List all knowledge bases with metadata and status.
Synchronously delete a document from qdrant, catalog and disk.
Semantic search over chunk bodies of a kb with metadata filters, embedding and rerank scores.
Document-level PubMed-style search with optional FTS query and metadata filters.
Parameter constraints missing. E.g., top_k, page, page_size lack min/max bounds. min_score lacks range. sort parameter in list_documents has no enum of valid values. LLMs will pass invalid values without guidance.
Error handling not documented. No guidance on retryability, user-fixable errors, or recovery steps. E.g., if a PDF path is invalid or a KB does not exist, the LLM has no recovery hint.