Claude Code now reports which MCP servers config validation dropped

Claude Code's headless runs now report which MCP servers config validation dropped. Version 2.1.219 adds mcp_server_errors to the stream-json init event, listing every --mcp-config entry that failed validation, and terminal runs get a startup warning for the same problem, per the changelog at code.claude.com.

Before this release, a malformed MCP config entry meant the agent ran to completion with fewer tools than you expected and nothing in the output told you why. That's a bad failure mode because it doesn't look like a config error. It looks like the model got worse at using tools, or forgot a capability it had yesterday. I've lost a full day chasing that exact symptom, checking prompts and model behavior before it ever occurred to me to check whether a server had silently dropped out.

Headless runs are the harder case here. There's no terminal to print a warning to, no human watching the session start up. If mcp_server_errors stays empty in your stream-json init event, that's now a real signal you can check in CI or in any script that shells out to Claude Code, instead of inferring tool loss from downstream behavior.

The same release adds HTTP status codes and error text to claude mcp list and /mcp when a connection fails, and a separate warning when a config value has hidden leading or trailing whitespace, another classic silent-failure source. None of these are big features. They're the kind of observability gap that's invisible until you've burned hours misdiagnosing it as something else.

Watch whether mcp_server_errors becomes a standard check in CI setups that run Claude Code headlessly. If a bad config can now fail loud instead of failing quiet, that's the whole fix.