{"version":"2.1.122","anchor":"mcp-long-running-task-tracking","canonical_anchor":"mcp-long-running-task-tracking","heading":"MCP Long-Running Task Tracking","tier":null,"area":null,"scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.122\/e\/mcp-long-running-task-tracking","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.122","markdown":"### MCP Long-Running Task Tracking\n\n**What**\n\nAdds infrastructure to track, poll, persist, and resume MCP tasks that run asynchronously on MCP servers using the experimental tasks API.\n\n**Details**\n\n- When a tool call spawns an MCP task, Claude Code now registers it in a task registry and polls the server for status updates\n- Task metadata is persisted to disk at `<session-dir>\/mcp-tasks\/mcp-task-<id>.meta.json` so tasks survive session restarts\n- On resume, previously-running MCP tasks are reconnected to their servers and polling resumes\n- Task status updates (completed, failed, cancelled) are injected back into the conversation as task-notification messages\n- The status bar shows `1 MCP task` (or similar) during active polling\n- Supports kill\/cancel from the client side\n\n**Evidence**\n\nFull task lifecycle management (search for `\"mcp_task\"`, `\"restoreMcpTasks\"`, `\"removeMcpTaskMetadata\"`, `\"Task was cancelled by the server.\"`)"}