{"version":"2.1.295","anchor":"mcp-servers-that-failed-are-re-dialed-at-the-start-of-the-ne","canonical_anchor":"mcp-servers-that-failed-are-re-dialed-at-the-start-of-the-ne","heading":"MCP servers that failed to connect can be retried at the start of the next turn","tier":"soon","area":"MCP","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.295\/e\/mcp-servers-that-failed-are-re-dialed-at-the-start-of-the-ne","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.295","markdown":"### MCP servers that failed to connect can be retried at the start of the next turn\n\nMCP servers that failed with a retryable error can be retried automatically at the start of each turn, for a limited number of turns\n\n**What**\n\nMCP servers are outside programs that give Claude extra tools. A turn is one round of you sending a message and Claude answering. When an MCP server fails to connect with an error that is worth retrying, Claude Code can now try to connect to it again at the start of each following turn.\n\n- Each retry has a time limit. A retry that takes too long is given up on.\n\n- The number of turns a server is retried for is set remotely, and is capped at 10.\n\n- If the server sends no value, the number is 0 and no retries happen. The feature is off unless it is switched on remotely.\n\n**Why**\n\nWhere it is switched on, a server that failed briefly can come back by itself, without you running `\/mcp` to reconnect it by hand.\n\n- Flag `tengu_snug_reef`: Not enough to say (read for one account on one subscription tier against v2.1.295; this account: no value returned, anonymous baseline: no value returned, compiled default: not a boolean we can read) These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.\n- Area: MCP\n- Tier: Nothing to try yet\n- Useful: 3\/5\n- Signal: 3\/5\n- Scope: individual\n- Heads-up: no\n- Present in the build but not switched on"}