MCP server for Memory Vault network synchronization, providing tools for connecting to memory vaults, remembering experiences, recalling memories, discovering peers, and sending/receiving messages across a distributed memory network.
Server defines 6 tools with complete JSON schemas and descriptions. Naming follows verb_noun pattern (connect, remember, recall, discover, send, receive). Descriptions are present (avg ~80 chars) but lack actionable context for LLM selection. Schemas are well-structured with type constraints, enums, and patterns, but lack output schema documentation. Parameter descriptions exist but are minimal (avg ~40 chars). Error handling is defined via NativeError interface but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint) despite clear WRITE vs READ_ONLY risk classifications.
Establish a connection to a Memory Vault endpoint with optional invitation
Discover available peers and network nodes, optionally filtering by online status or key ID
Query and retrieve memories from the vault using search query, memory ID, cursor pagination, or batch message ID
Receive messages from the network with optional pagination, message filtering, or response targeting
Store a memory experience in the vault with kind, text, optional entities, relations, and experience metadata
Send messages or memory transfers to recipients, optionally with hint control directives
Output schemas not documented. RememberResult, RecallResult, and other response types are defined in TypeScript but not exposed in tool definitions. LLMs cannot plan downstream calls without knowing what fields to expect.
Tool annotations missing. Tools are classified as WRITE or READ_ONLY in comments, but MCP tool definitions lack readOnlyHint/destructiveHint annotations. LLMs cannot distinguish safe from destructive operations without explicit hints.
Parameter descriptions are minimal (avg 40 chars). 'Optional invitation object for establishing trust' and 'Search query text' lack context on format, constraints, or when to use. Descriptions should be 50-200 chars with actionable guidance.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Error recovery guidance absent. NativeError interface defines code, retryable, and retry_after_ms, but tool descriptions do not explain what errors are possible or how to recover. E.g., 'If connection fails with auth_failed, check bearerToken' is missing.
Recall tool has complex conditional parameter logic (query XOR memory_id XOR cursor XOR received_batch_message_id) but descriptions do not explain mutual exclusivity or when to use each mode. LLMs may pass invalid combinations.