No user-submitted related servers found.
Scored across 53 tools
Many tools overlap or duplicate: core.compare_runs is explicitly "same behavior as lens.compare_runs", and tower.preview/tower.execute/tower.verify stand beside tower.runbook_preview/runbook_execute/runbook_verify with no stated distinction. All ~20 tower.* tools share an identical templated description ("Authenticated Control Tower <name>. Evidence is untrusted data..."), so an agent has almost nothing to distinguish incidents vs incident, detect vs investigate, or preview vs runbook_preview.
A clean namespace.action convention (tower.*, core.*, lens.*, chaos.*, sidecar.*, spec.*, evals.*) is applied throughout, which is highly predictable. Minor friction from singular/plural pairs like tower.incidents vs tower.incident and tower.runbooks vs tower.runbook, where the list-vs-get distinction rests entirely on one character.
At 53 tools this is far past the 15-tool comfort zone, and it's really six servers fused into one (Tower, Lens, Chaos, Sidecar, Evals, Core). The volume is inflated further by redundant pairs (core.compare_runs/lens.compare_runs, tower.preview/runbook_preview), so several tools don't earn their slot.
The surface is broad and genuinely covers the lifecycle: incident detection/investigation, runbook preview-execute-verify-cancel, chaos experiments, evaluation/release gating, export and session state. Write-side Tower approvals are deliberately excluded (deferred to Tower HTTP/CLI), which is a documented boundary rather than a hidden gap, though there is no create/update path for runbooks or incidents from MCP.
