MCP server for enabling memory for AI assistants through a knowledge graph
Strong foundation with complete schemas, good descriptions, and proper tool annotations. All 10 tools have clear names starting with action verbs (create_, delete_, read_, search_, update_, open_), descriptions (avg ~80 chars), and documented input/output schemas using Zod. Tool composition is sound, each tool has one responsibility. However, parameter descriptions lack depth (missing format constraints, ranges, examples of valid values), error handling guidance is absent, and output schemas omit pagination/limit details for read operations. Batch operations are well-designed but could benefit from clearer guidance on result limits and field mappings.
Add new observations to existing entities in the knowledge graph
Create multiple new entities in the knowledge graph
Create multiple new relations between entities in the knowledge graph. Relations should be in active voice
Delete multiple entities and their associated relations from the knowledge graph
Delete specific observations from entities in the knowledge graph
Delete specific relations from the knowledge graph
Open a set of entities and return their relations and neighbors
Query parameter in read_graph and search_nodes lacks format constraints. Description says 'Optional search query' but does not specify: is it free-text, regex, SQL-like syntax? What characters are allowed? What is the max length? LLMs will guess and pass invalid queries.
No error handling guidance in any tool description. If create_entities fails because an entity name already exists, or a relation references non-existent entities, what should the LLM do? Descriptions do not say 'retryable', 'user-fixable', or 'fatal'. Agents cannot plan recovery.
read_graph and search_nodes return unbounded result sets. No pagination, limit, or offset parameters. Description does not state a result cap. If the knowledge graph has 10,000 entities, both tools return all of them, blowing the context window.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2025-06-18+ | v2 |
Read the full knowledge graph or search for entities and relations by name, entity type, or observation content
Search for entities and relations by name, entity type, or observation content
Update observations for entities in the knowledge graph
Destructive tools (delete_entities, delete_observations, delete_relations) lack confirmation or dry-run support. An agent can irreversibly delete the entire knowledge graph in one call. No pattern:confirmation-request or rollback guidance.
Parameter descriptions are minimal. 'The name of the entity' does not explain: is it case-sensitive? Max length? Allowed characters? Must be unique? LLMs cannot infer these constraints and will pass invalid values.