Search is plain text matching over the local store, which means results are predictable and immediate — you get the sessions that literally contain the phrase, rather than whatever an embedding model thought was close. The archive side is the other half of the value and the part to handle carefully: back up before restoring, since restore overwrites current history and migrating in move mode deletes the original session.
A reader and archive tool for Cursor's local AI chat history. It reads Cursor's SQLite store directly — no vectors, no embedding model, no separate service — and can also move, back up and restore that history.
- cursor_history_search does grep-style matching across every session and returns each hit with surrounding context
- cursor_history_list shows recent sessions with their workspace, message count and timestamps, paged with limit and offset
- cursor_history_show returns a whole session — messages, tool calls and AI responses
- cursor_history_export writes a session out as Markdown or JSON
- cursor_history_backup saves a portable archive of all chat history; cursor_history_restore reads one back, and says plainly that it overwrites what is there now
- cursor_history_migrate moves or copies sessions between workspaces, deleting the original when moving rather than copying
- cursor_history_year_pack builds a sanitised year-in-review summary — statistics, topics and keywords — plus a prompt template for turning it into a report, and is read-only
Cursor installed with existing chat history, and Node.js 20+. It runs as npx cursor-history-mcp — nothing to install, no account, no key, and it works offline since everything is read from the local database. Note that restore and migrate change that history: restore overwrites it, and a move deletes the source session. MIT.
One command — npx -y cursor-history-mcp