MCP server for ontological graph-based RAG and requirement recommendation using Neo4j knowledge graph
14 tools with READ_ONLY and WRITE risk classifications. Tool names follow verb_noun pattern (list_, search_, get_, run_, record_) which is good. However, descriptions are minimal (10-60 chars), parameter descriptions are sparse or absent, and input schemas lack proper type definitions and constraints. No output schemas documented. Error handling guidance absent. Most tools appear to be thin wrappers around Neo4j/document APIs without LLM-optimized descriptions or recovery guidance. Composition is reasonable (single responsibility per tool), but parameter naming lacks consistency (e.g., 'path' vs 'project_id' vs 'adr_id', no suffix convention). No evidence of pagination, limits, or structured output shaping.
Generate a report of extraction results
Get all requirements for a specific ADR (Australian Design Rule)
Identify compliance gaps between requirements and project assignments
Get detailed information about a specific project
List all projects in the Neo4j knowledge graph
List all indexed documents in the configured document roots
Execute a read-only Cypher query against the Neo4j knowledge graph
No input schemas visible for 8 tools (list_documents, read_document, summarize_document, list_all_projects, extraction_report).
Descriptions are 10 - 60 characters (well below 194-char baseline). Too short to guide LLM selection. E.g., 'List all indexed documents in the configured document roots' (60 chars) lacks context on when to call vs search_local_documents, what structure is returned, or prerequisites.
Parameter descriptions missing or minimal. 'path' parameter in read_document has no description of expected format (absolute? relative? file extensions?). 'query' in search_local_documents lacks guidance on syntax, supported operators, or field scope. LLMs cannot infer constraints from names alone.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | <=2025-11-25 | v2 |
Read the full content of a document
Record user feedback on a requirement-project link for reinforcement learning
Run PDF/document extraction to identify requirements
Search for text in local documents (CSV, PDF, DOCX, TXT, MD files)
Suggest document sources for a requirement based on semantic similarity
Summarize the content of a document
Record an inferred link between a requirement and project with confidence score
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. E.g., does list_documents return [name, path, size, modified_date]? Does get_project return team_id needed for chaining? Unknown.
No error handling guidance. Tools lack recovery hints (e.g., 'If document not found, try search_local_documents()'). No categorization of errors as retryable vs user-fixable. Agents cannot self-correct on failures.