Is CLAUDE.md enough for persistent memory?

CLAUDE.md is excellent for durable project instructions loaded every session. It does not replace a shared, searchable store of living decisions and open tasks that Cursor and cloud agents can also write. Use both: CLAUDE.md in git for standards; ReadmeDB for the cross-tool project brain.

Does Claude auto memory sync to Cursor?

No. Claude Code auto memory lives under Claude’s local project memory paths. Cursor cannot read it. ReadmeDB stores human-readable markdown both MCP clients can call.

readmedbmcp

Solution · Claude Code

Claude Code persistent memory — beyond CLAUDE.md alone

CLAUDE.md and auto memory keep one tool, one machine in sync — but the moment you switch to Cursor, or open the same repo on another machine, that truth doesn't follow. ReadmeDB is the shared project brain Claude Code and Cursor both read and write via MCP, so decisions don't diverge across tools.

What Anthropic already gives you

CLAUDE.md

Project (or user) markdown Claude loads at session start — architecture, commands, conventions. Commit it to git so humans and other tools can read the static briefing.

Auto memory

Notes Claude writes while you work (preferences, project facts). Powerful inside Claude Code — invisible to Cursor, Codex, and teammates’ agents.

Where local memory breaks

You switch to Cursor and the auto-memory trail is gone
A cloud agent cannot see ~/.claude/projects/…/memory/
Living open tasks and mid-sprint decisions rot in chat instead of a shared file

Use both: CLAUDE.md + ReadmeDB brain

Keep standards in CLAUDE.md. Put the living brain in ReadmeDB:

decisions.md — ADR-style choices agents must not re-litigate
open-tasks.md — handoff for the next session or tool
preferences.md — personal or team “never do X” rules

Optional: add one line in CLAUDE.md telling Claude to call get_namespace_summary / read those files at session start.

MCP config for Claude Code

{
  "mcpServers": {
    "readmedb": {
      "url": "https://app.readmedb.com/api/mcp",
      "headers": {
        "Authorization": "Bearer rdb_your_key"
      }
    }
  }
}

Same block works in Cursor. One API key = one namespace = one project brain. Free tier available; Pro is £5/mo when you outgrow it.

Start free — Claude + Cursor shared brain →

Cursor setup AGENTS.md vs CLAUDE.md Codex Project brain For teams vs Mem0 MCP file storage