Description

Claude Code v2.1.117 (April 22, 2026) and v2.1.119 now connect MCP servers in parallel on startup and during subagent/SDK MCP server reconfiguration. Zeph's current MCP lifecycle in zeph-mcp connects servers sequentially.

Impact

For users with 3+ MCP servers configured, parallel connection reduces startup latency proportionally. With typical 500–1500ms handshake per server, 4 servers = 2–6s sequential vs ~500ms parallel.

Relevance to Zeph

zeph-mcp/src/lifecycle.rs connects and initializes servers in a loop. Converting to FuturesUnordered or join_all would give the same benefit with minimal code change.

Implementation Sketch

• Replace sequential server init loop with futures::future::join_all or tokio::task::JoinSet
• Per-server errors should be non-fatal (log and skip, don't abort remaining)
• Startup spinner in TUI should aggregate per-server status (already has system status)

Estimated Complexity

Low (~2h). Pattern is straightforward async parallelization.

Source: Claude Code changelog v2.1.117 — https://code.claude.com/docs/en/changelog