No user-submitted related servers found.
Scored across 18 tools
The memory_* and episode_* prefixes cleanly separate two tiers (durable memories vs. journal entries), and within each tier the operation verbs (search/list/show vs. write/update/remove/verify) are clearly distinct. The confirm/cancel and scope_disable/enable pairs are unambiguous complements. Even the two search tools are well-differentiated: memory_search is ranked while episode_search is explicitly chronological.
All 18 tools follow a strict resource_prefix + action_verb pattern in snake_case: memory_search, memory_write, episode_handoff, memory_scope_disable. Compound names like memory_write_confirm and memory_scope_overview extend the pattern predictably. No camelCase, no vague verbs, no mixed conventions.
18 tools is slightly above the ideal 3-15 band, but the domain is genuinely rich — memory lifecycle, verification, drift detection, scoping, episodes, and audit all earn dedicated tools. One or two (memory_audit_turn, which is hook-dispatched) could arguably be trimmed from the model-facing surface, but none feels redundant.
The core CRUD lifecycle is well covered (write/update/remove/show/verify/search), but the tool descriptions repeatedly reference tools that are absent from the exposed surface: memory_restore, memory_list_tombstones, memory_health, memory_proposals, and memory_conflicts are all invoked as if callable but not listed. An agent following the memory_remove description's 'use memory_restore(id) to undo' instruction would hit a dead end. There is also no episode_update or episode_remove, leaving the journal tier with no correction path other than promotion.
