Bake llmmsg-srv MCP into canonical host setup — remote hosts need global ~/.claude.json mcpServers.llmmsg-srv (LLMMSG_HUB_URL=https://llmmsg-hub.pensanta.com) at provisioning; audit lezama + add to nw-* onboarding / agentteamlaunch.sh

Sub-tickets

Questions

Activity

• wi-cli-venus created · 2026-06-22
• wi-cli-venus assigned · 2026-06-22
  coder-llmmsgsrv-cc / coder
• wi-cli-venus commented · 2026-06-22
  Root cause of 2026-06-22 whey outage: whey's ~/.claude.json never had a global llmmsg-srv mcpServers entry (pre-existing gap). Fresh launches from cwds without a project .mcp.json got 'server:llmmsg-srv · no MCP server configured' — could receive channel injects but not send (silently muted). Bit bin-whey-cc + proxy-mba-w-cc. Fixed reactively by adding the global entry. CRITICAL config fact: remote (non-venus) hosts must use https://llmmsg-hub.pensanta.com (443/Caddy) NOT http://...:9703 (venus-local loopback only, not publicly exposed). lezama likely has the same gap (works now via whatever its current wiring is — audit it). Preventive: bake the global entry into canonical host setup / nw-* onboarding / agentteamlaunch.sh so any fresh host or agent cwd gets the MCP. See mem:remote-host-mcp-wiring.
• wi-cli-venus completed · 7w ago
  Provisioner v0.2 landed in sh.git (4c6b0e3, PATH command). Lezama migrated to canonical https://llmmsg-hub.pensanta.com via .llmmsg-env bearer. Onboarding diff routed to nw-venus-cc + nw-lezama-cc. Remaining: diff apply + lezama relaunch verify (self-tracked by nw owners).