Modular Retrieval Augmented Generation (RAG) MCP Server - A pluggable RAG service with hybrid search (dense + sparse), document ingestion, and knowledge base querying capabilities
The server defines 3 tools with READ_ONLY risk classification, but exhibits significant gaps in schema completeness, parameter descriptions, and output documentation. While tool names follow verb_noun conventions (query_*, list_*, get_*), the visible schema definitions lack comprehensive parameter descriptions, and there is no evidence of output schemas or error handling guidance. The server is early-stage (v0.1.0) and shows modularity in design, but does not meet production quality baselines for agent tool definition.
Get a summary of a specific document in the knowledge hub
List all available document collections in the knowledge hub
Query the knowledge hub using hybrid search (dense + sparse retrieval with optional reranking)
No output schemas documented for any tool. Agents cannot plan downstream actions or validate returned data without knowing field names, types, and structures.
Parameter descriptions are minimal or missing constraints. 'top_k' (integer) lacks min/max range; 'collection' parameter does not explain format or valid values; 'document_id' does not specify ID format or how to obtain valid IDs.
Tool descriptions lack context on when to use vs alternatives, expected return format, and prerequisites. E.g., query_knowledge_hub does not state result count, pagination support, or whether results are ranked by relevance.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 40 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 33 | - | v1 |
No error handling or recovery guidance. No documentation of when tools fail (e.g., invalid collection, document not found, search timeout) or what the agent should do next.
Tool chaining is unclear. query_knowledge_hub returns results, but there is no evidence that returned document_ids can be passed directly to get_document_summary, or that list_collections returns collection names usable in query_knowledge_hub.
Parameter 'enable_rerank' defaults to true with no documentation of performance impact, when to disable, or what reranking algorithm is used. Agents may not understand when to set it to false.