Scored across 33 tools
Most tools have a distinct purpose, but several pairs overlap: perseus_health/perseus_get_health, local vs external memory (perseus_memory/perseus_mneme/perseus_mimir alias), and perseus_tokens/perseus_budget. Descriptions usually clarify the boundary, but an agent could still mis-select between related recall, health, and token-budget tools.
All tools share the perseus_ prefix and lowercase style, which is helpful, but the second half mixes nouns (tokens, health, profile), verbs (read, include, capture), and quirky names (perseus_perseus, tooltrim, get_health vs health). The pattern is readable but not consistent enough to predict tool names.
33 tools is well beyond the comfortable 3-15 range and even past the 25-tool threshold for 'too many'. Many are small directives or niche utilities that could be consolidated; the count dilutes the server's coherence.
The core context lifecycle is well covered: render (get_context/include/read), configure (profile/prompt/budget), recall (memory/session/waypoint), and health-check (health/drift). Notable gaps are mostly read-only surfaces like tasks, inbox, and skills, which have no write/manage counterparts, but these feel outside the server's primary context-rendering purpose.
