{"version":"2.1.300","anchor":"mcp-status-prompt-now-distinguishes-retrying-from-failed-for","canonical_anchor":"mcp-status-prompt-now-distinguishes-retrying-from-failed-for","heading":"Claude now hears whether a broken MCP server is retrying or out for the session","tier":"notice","area":"MCP","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.300\/e\/mcp-status-prompt-now-distinguishes-retrying-from-failed-for","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.300","markdown":"### Claude now hears whether a broken MCP server is retrying or out for the session\n\nThe status note about broken MCP servers now says whether a server needs sign-in, is retrying, or has failed for the rest of the session\n\n**What**\n\nMCP servers are outside programs that give Claude Code extra tools. When one of them is not working, Claude Code adds a status note to the conversation for the model. That note now puts each such server into one of three states:\n\n- `needs_auth`: the server needs you to sign in.\n\n- `failed_retrying`: the server failed and Claude Code is still trying to reconnect.\n\n- `failed_for_session`: the server failed and will stay unavailable until the session ends. A session is one run of Claude Code from start to exit.\n\nBefore, all blocked or failed servers went into one list with no difference made between them. Claude Code also now keeps track of servers whose \"comms\" role is turned off and takes them into account when it works out which tools are available. Headless runs, where Claude Code works without the interactive screen, now keep a record of disabled MCP servers.\n\n**Why**\n\nThe model can tell whether a tool from a broken server may come back soon or will not return this session. It can wait for a reconnect in the first case and move on in the second.\n\n- Area: MCP\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 1\/5\n- Scope: individual\n- Heads-up: no"}