mcp-server-recovery-restart-not-required
Failed MCP servers (researchclaw, xint) were recovered within the same session by fixing root causes (Traefik network label, transport type re-registration) without requiring a claude-code session restart. Changes to MCP server infrastructure (docker-compose labels, container recreation) are reflected immediately when the MCP client retries connection. Session restart is only needed when changing the MCP registration itself in .claude.json.