MCP server for Hydra DB: State-of-the-art agentic memory with recall, ingest, and knowledge graph context
The HydraDB MCP server demonstrates reasonable structure with 14 canonical + 7 deprecated tools. Tool names follow verb_noun convention (query, ingest, list, inspect, delete, etc.), which is solid. However, parameter descriptions are minimal or absent across most tools, schemas lack type information in many cases, and output schemas are entirely undocumented. The 7 deprecated aliases are properly marked but still registered, creating surface area bloat. Error handling and security considerations are not visible in the provided code. The server benefits from clear naming and a logical tool taxonomy (canonical, OAuth-gated, BYOG graph families), but falls short on LLM-critical details: actionable parameter guidance, output structure documentation, and recovery hints.
DEPRECATED: Use hydradb_delete instead
DEPRECATED: Use hydradb_inspect instead
DEPRECATED: Use hydradb_ingest instead
DEPRECATED: Use hydradb_list instead
DEPRECATED: Use hydradb_list instead
DEPRECATED: Use hydradb_query instead
DEPRECATED: Use hydradb_ingest instead
Output schemas completely undocumented. No visible schema definitions for any tool's return value. LLMs cannot plan downstream calls or extract result fields without knowing what structure to expect.
Parameter descriptions are minimal or absent. E.g., hydradb_status takes no parameters and has empty input schema. hydradb_databases, hydradb_list_collections, hydradb_graph_collections have no input schema visible. LLMs cannot understand what values to pass or when constraints apply.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2025-06-18+ | v2 |
List available databases this connection can address (OAuth only)
Delete a memory or knowledge source from HydraDB
Delete a collection
Provide feedback about a query that already ran
Administrative operations on BYOG graph collections
List available BYOG graph collections
Query BYOG graph collections using Cypher
Ingest memories and documents into HydraDB
Inspect details of a specific memory or knowledge source
List memories and knowledge sources from HydraDB
List collections in the current database
Query memories and knowledge sources in HydraDB
Get the status of HydraDB connection and ingestion
Get the connected subgraph of one item - what it links to, what shares its thread, parent/child relationships
7 deprecated tools (hydra_db_search, hydra_db_store, hydra_db_ingest_conversation, hydra_db_list_memories, hydra_db_list_sources, hydra_db_fetch_content, hydra_db_delete_memory) are still registered with empty input schemas and minimal descriptions marked DEPRECATED. These create surface bloat, increase LLM confusion, and should be removed or hidden from the tool list in production. Keeping them through a major version boundary is reasonable, but they should not be advertised to new agents.
Destructive operations (hydradb_delete, hydradb_delete_collection) have no documented confirmation or dry-run capability. Descriptions do not warn that deletion is irreversible or suggest verification steps. No error handling guidance visible.
Tool descriptions lack actionable context. E.g., hydradb_query says 'Query memories and knowledge sources' but does not explain when to use it vs hydradb_list, how ranking works, what to do if no results match, or what fields the response includes. Descriptions should be 50 - 200 chars with WHAT, WHEN, and HOW context.
Parameter 'kind' in hydradb_query and hydradb_list is an enum with valid values ('memory', 'knowledge', 'all' for query; 'memory', 'knowledge' for list), but descriptions do not explain the semantic difference. When should an agent pick 'memory' vs 'knowledge'? What data lives in each?
hydradb_ingest has a 'turns' parameter (array) with no description. What structure should turns have? Is it optional? What happens if omitted? Undocumented array parameters invite malformed calls.
hydradb_graph_admin 'operation' parameter is a free-form string with no enum or constraint. What operations are valid? Examples: 'create', 'delete', 'update'? Without constraints, LLMs will hallucinate values.
No error handling guidance visible in tool definitions. How should an agent respond if hydradb_query returns no results? If hydradb_delete fails (e.g., item not found)? If hydradb_ingest hits a rate limit? Error messages should guide recovery (retry, user input, abort).
No pagination guidance documented for list-style tools. Does hydradb_list respect a 'limit' parameter to constrain results? What is the default limit? If >1000 memories exist, how does the agent retrieve all of them? Pagination support and constraints should be explicit.