Cursor persistent context — without re-explaining every chat

Deep debugging in Cursor, then switch to Claude Code and spend ten minutes re-briefing. ReadmeDB is the shared project brain both tools write to — versioned markdown files, not another siloed chat memory.

What Cursor already gives you (and where it stops)

Cursor rules (.cursor/rules, project rules) are great for stable conventions on one machine. They do not:

• Survive as a living handoff when you open Claude Code on another box
• Capture “what we decided yesterday” as searchable, versioned files
• Let a teammate’s agent read the same open-tasks.md you just updated

The shared project brain pattern

Keep durable truth in markdown your agents own:

• decisions.md — why Postgres, why not Redis for X
• open-tasks.md — what the last Cursor session left unfinished
• preferences.md — stack, style, “never touch billing”

Cursor writes them via MCP. Claude Code reads them via the same key. You stop pasting context between tools.

Connect Cursor in 30 seconds

Add this to .cursor/mcp.json (or Cursor Settings → MCP):

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

Free plan: 50 files · 5 MB · 3 keys. No card. Then ask Cursor: “Create decisions.md and open-tasks.md in ReadmeDB from what we know about this repo.”