# Search sessions

> Deduplication across related searches.

Canonical: https://keta1930.github.io/icml2026-mcp/en/docs/concepts/sessions



A search session remembers which papers it has returned. Reusing its `session_id` suppresses those papers in later searches, including papers recorded through `get_papers`.

Sessions are optional. Calls without a `session_id` are stateless, and different sessions do not share state.

The session ID is a private capability. The server stores only its hash and cannot list or recover raw IDs. Sessions expire after inactivity and have bounded rounds and result counts.