Related Servers

Many tools are near-duplicates separated only by authority prefix: task_archive vs manager_task_archive vs dashboard_task_archive, task_supersede vs manager_task_supersede, needfix_update vs dashboard_needfix_update, task_show vs dashboard_task_detail vs agent_task_status, task_health vs dashboard_health vs dashboard_snapshot. The descriptions do disambiguate with 'use X instead of Y' prose, but with 213 tools sharing overlapping verbs and resources, an agent must read long docstrings to pick correctly.

The set mixes three schemes: prefixed families (aiworkhub_task_*, aiworkhub_manager_*, aiworkhub_dashboard_*, aiworkhub_agent_*, aiworkhub_quality_*), bare families (needfix_*, roadmap_*), and inconsistent suffixes like _readonly, _readonly-only variants, and _helper terms. Verb ordering is usually verb_noun but there are noun-first outliers (task_health, dashboard_snapshot, completion_inbox, review_packet, quality_profile).

213 tools vastly exceeds any reasonable scope for a single MCP surface, and many are parallel re-exports of the same operation under different authority prefixes. The count is an extreme mismatch for coherent agent tool selection.

The domain surface is extensive: task lifecycle, NeedFix, roadmap, skills, recipes, SDLC cases, storage retention, dashboard reads, launch/dispatcher control, and quality evidence all have create/read/update/archive paths. Coverage is close to complete, though a few lifecycle edges (e.g. roadmap delete/archive, consistent unlock of quarantined artifacts) are implicit rather than explicit.