mcp-session-lifecycle

The MCP server manages notebook session lifecycle using a proxy supervision layer that spawns child processes, detects binary version changes, and seeds rejoin environment variables to restore sessions after daemon restarts. A daemon watch loop continuously classifies and acts on rejoin decisions.

Race conditions between background rejoin and user tool calls in MCP occur when concurrent operations attempt to modify the shared notebook session state simultaneously. The session state layer uses Arc<RwLock<Option<NotebookSession>>> with rejoin guards to prevent background rejoin from overwriting active user-initiated sessions.

Debug MCP session state and reconnection logic by examining the three-layer architecture: the proxy layer for child process spawning, the session state layer for Arc<RwLock> lock contention, and the daemon room layer for peer-based eviction timing after configurable windows with no peers.

The MCP daemon evicts rooms by UUID after a configurable window passes with no connected peers. Room state management is handled by the daemon room layer, which tracks peer presence and cleans up idle sessions to maintain system stability across the session lifecycle.

The require_handle pattern in MCP session access is a concurrency strategy used to minimize lock contention when accessing the notebook session. It ensures efficient read access to the shared session state without holding exclusive locks that would block rejoin operations.

Yes, the MCP proxy detects binary version changes during session reconnection. When an upgrade occurs, the proxy layer restarts the child process and seeds the NTERACT_MCP_REJOIN_NOTEBOOK environment variable to guide the reconnection logic back to the active notebook session.