{"version":"2.1.286","anchor":"mcp-prior-verdict-handling-made-more-resilient","canonical_anchor":"mcp-prior-verdict-handling-made-more-resilient","heading":"MCP servers reconnect more reliably after a failed tool listing","tier":"notice","area":"MCP","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.286\/e\/mcp-prior-verdict-handling-made-more-resilient","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.286","markdown":"### MCP servers reconnect more reliably after a failed tool listing\n\nClaude Code now forgets a stale MCP connection result when listing a server's tools fails, and logs a restarting server's error output\n\n**What**\n\nMCP (Model Context Protocol) servers are outside programs that give Claude extra tools. Claude Code changes how it connects to them:\n\n- If asking a server for its tools times out or is refused, Claude Code forgets what it had remembered about which protocol version the server supports, and checks again on the next connection.\n\n- A rejection that arrives after the connection has been set up is now handled as its own case.\n\n- When a server running as a local process restarts, Claude Code logs that server's error output.\n\n- Connection records now use the server's real connection type instead of always marking it as a proxy connection.\n\n**Why**\n\nA server that failed once is less likely to stay stuck on an outdated result. The logged error output also makes a server that keeps restarting easier to diagnose.\n\n- Area: MCP\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 1\/5"}