MCP server for intelligent knowledge base search and retrieval with Dify integration, providing comprehensive search capabilities including keyword generation, file searching, content boosting, and answer extraction.
The server defines 5 tools with complete input schemas and descriptions. However, quality is mixed: naming conventions are inconsistent (verb-noun not uniformly applied; 'file_discover' lacks clarity vs 'file_lister'), descriptions are present but brief (average ~60 chars, below the 194-char baseline for A+ tools), and critically, NO tool provides documented output schemas, LLMs cannot predict what fields to expect or chain results to downstream tools. Parameters are well-typed with constraints (enums, defaults, ranges), but descriptions lack specificity about when to use each tool and how they differ. No error handling guidance, no recovery patterns, and no field-naming alignment checks between tools (e.g., resource_id is used across tools but not validated for consistency). This is a functional server for discovery and retrieval tasks, but falls short of production-grade clarity.
Search and extract answers from knowledge bases.
Discover files using backend retriever and DSPy.
List files in knowledge base resource with pagination support.
Generate keyword sets for search.
Retrieve information from knowledge base.
No output schemas documented. LLMs cannot predict return types, field names, or structure for any of the 5 tools. This prevents downstream tool chaining and forces agents to guess at field names (e.g., does retriever return 'documents' or 'results' or 'matches'?). Breaks pattern:tool-chain and mxe:include-chaining-ids.
Inconsistent tool naming reduces clarity. 'file_discover' is ambiguous (discover what? files by relevance? by name?), while 'file_lister' clearly lists. 'keyword_generator' starts with a verb, but 'file_discover' does not consistently follow verb_noun convention. Violates pattern:tool naming rule 'start with verb that reflects action.'
Tool descriptions are too brief and lack context for LLM selection. 'assistant' (60 chars) does not explain when to use it vs 'retriever' (71 chars). Below 194-char A+ baseline. No WHEN guidance, no prerequisites documented. LLMs must guess which tool to call.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-21 | D | 54 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 19 | - | v1 |
Parameter descriptions lack specificity about valid values. 'search_method' (retriever) lists options in description ('hybrid_search, semantic_search, full_text_search, keyword_search') but should use an enum constraint in the schema. 'backend_type' (file_discover, file_lister, retriever) accepts free-form string with only example 'dify' in description, no enum boundary. LLMs may hallucinate invalid backend types.
No error handling or recovery guidance. If 'resource_id' does not exist, if 'query' returns no results, or if 'backend_type' is invalid, tools provide no guidance for the agent (no retry heuristics, no alternatives, no actionable error messages). Violates pattern:recovery-guide and pattern:error-classification.
Tool relationships not documented. 'file_discover' and 'retriever' both search knowledge bases, no description explains the difference (file_discover = file-level relevance; retriever = passage/chunk-level?). LLM cannot determine when to use each. Violates review:name-clarity and mxe:include-chaining-ids.
Parameter dependencies not documented. 'reflection_threshold' and 'enable_reflection' in assistant are coupled, but description does not say 'threshold only applies if enable_reflection=true.' Similarly, 'file_discovery_evaluation_threshold' in assistant depends on 'enable_file_discovery_evaluation', no explicit dependency noted. Violates review:param-relationships.
Numeric parameters lack explicit bounds. 'top_k_recall' (file_discover, default 100) and 'top_k' (retriever, default 40) have no documented max. Can an agent pass top_k=10000 without breaking performance? 'limit' (file_lister) lacks min/max. 'max_sets' (keyword_generator) lacks bounds. Violates review:param-validation-rules.