Guide

One memory for Claude Code, Cursor and Codex

Teach Claude Code a project rule on Monday and Cursor still does not know it on Tuesday. Each tool keeps its own memory in its own format: CLAUDE.md and auto memory for Claude Code, rules files for Cursor, AGENTS.md for Codex. Switching tools means explaining the project again.

memgit runs one MCP memory server that every one of these tools connects to. A rule saved from any of them is there in all of them, in the next session.

Set it up

pip install memgit
memgit init

memgit init creates the store, imports the memory Claude Code already keeps, and offers to connect every AI tool it finds on the machine. To connect them all without the questions:

memgit setup all

Or connect one tool at a time: memgit setup claude-code, cursor, codex, windsurf, gemini-cli, cline, continue or claude-desktop. Each command writes the memgit entry into that tool's own MCP config and is safe to re-run. For any other MCP host, memgit setup print-config prints the snippet to paste.

Check it worked

Ask Claude Code to remember a rule, then open Cursor or Codex on the same project and ask what the rules are. From a terminal:

memgit search "the rule you saved"
memgit log -n 3

Rules every tool reads, even without MCP

Some instructions should load in every session whatever the tool does with MCP. memgit core keeps a short per-project operating guide, and memgit core sync writes it into each tool's own rules file, so Claude Code, Cursor and Codex all start from the same rules.

Cost

All of this is free, MIT licensed and local: no account, no telemetry. Using the same memory on a second machine, or sharing it with a team, is memgit Pro, with 14 days free that start when the second machine connects. See memory on every machine.