{"version":"2.1.302","anchor":"mcp-task-polling-can-let-go-and-keep-polling-slowly-after","canonical_anchor":"mcp-task-polling-can-let-go-and-keep-polling-slowly-after","heading":"MCP tasks can keep being checked slowly after Claude Code gives up on them","tier":"notice","area":"MCP","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.302\/e\/mcp-task-polling-can-let-go-and-keep-polling-slowly-after","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.302","markdown":"### MCP tasks can keep being checked slowly after Claude Code gives up on them\n\nA long MCP task that would have failed can now be let go but still checked slowly, and picked up again when its server answers\n\n**Unclear.** Whether this mode is on by default is not settled.\n\n**What**\n\nAn MCP server is a separate program that gives Claude Code extra tools, and some of its work runs as long tasks that Claude Code checks on (polls) until they finish. Before, any of the following ended the task as failed:\n\n- the server is not configured\n\n- an identity mismatch\n\n- an era mismatch\n\n- the server needs you to sign in\n\n- reconnecting failed\n\n- the task's time limit expired\n\n- the server no longer offers the needed capability\n\nIn a new mode, Claude Code instead records the outcome, stops holding the process open for the task, and keeps checking on it at a slower rate. When the server answers again, the task is picked up and checked at the server's own pace.\n\n**Why**\n\nLong-running MCP tasks can survive a server outage or sign-in problem instead of being marked failed.\n\n- Area: MCP\n- Tier: You'll notice\n- Useful: 2\/5\n- Signal: 2\/5\n- Scope: individual\n- Heads-up: no"}