← Blog
·6 min read

How to give Cursor real memory with an MCP server

Cursor doesn't remember anything between sessions. Wire up the dropdat MCP server and Cursor can recall every ChatGPT, Claude, and Gemini chat you've captured. Setup walkthrough + example prompts.

Cursor is fantastic at the edit-and-explain loop, but the agent forgets everything between sessions. You can paste the same project context into chat ten times a day and still rediscover the same gotchas tomorrow.

The fix is small and concrete: give Cursor an MCP server that owns your long-term memory. dropdat's MCP server lets Cursor read any capsule you've captured from any AI app — across all your projects.

What you get

Setup (60 seconds)

  1. Install the dropdat browser extension and capture at least one chat.
  2. Generate an API key at dropdat.app/api-keys (`dk_live_…`).
  3. In Cursor, open MCP settings and add the dropdat server.
{
  "mcpServers": {
    "dropdat": {
      "command": "npx",
      "args": ["-y", "@dropdat/mcp"],
      "env": { "DROPDAT_API_KEY": "dk_live_…" }
    }
  }
}

Example prompts

Why this works better than @-mentioning files

File context tells Cursor *what* the code is. Capsule recall tells Cursor *what you and another AI already decided*. That's a much higher-signal slice of context — and it's the part that usually lives in a Slack message or a ChatGPT tab you closed three days ago.

See the Cursor setup page