Cursor forgets every Composer session. Fix that with one file.

.cursor/rules is the right place for “always use our error helper” — standing instructions you author once. It is the wrong place for knowledge that accretes: which approaches you rejected and why, how your team names things, what the client asked for last month. That content changes daily and nobody is going to hand-edit a rules file to keep up.

With a memory server attached, the Agent writes those facts as they happen and retrieves them by meaning — “what did we decide about auth?” finds the OAuth-over-magic-links decision even though the word “auth” never appears in it.

Add the server — globally via Settings → MCP, or per-project. Cursor opens a browser tab the first time it connects, where you sign in and approve it; there is no key to create first, and you can revoke it later from the dashboard:

{
  "mcpServers": {
    "Kognite": {
      "url": "https://mcp.kognite.dev/mcp",
      "type": "streamable-http"
    }
  }
}

• ✓Toggle the Kognite server ON in Settings → MCP — a green dot means its tools are loaded.
• ✓Open Composer in Agent mode; the memory_* tools are now callable.
• ✓No plugin, no npm install, no local process — the URL is the whole integration.

Cursor uses tools on its own judgment in Agent mode. One small rules file makes remembering and recalling the default instead of the exception:

Before answering anything that depends on project history or my
preferences, call memory_search first. When a durable fact, decision,
or preference is stated, persist it with memory_add.

From then on the loop closes itself: decisions made in one Composer session are context in the next one.

• ✓Cross-session continuity — the Agent recalls prior decisions instead of re-deriving (or contradicting) them.
• ✓Cross-tool continuity — the same store serves Claude Code, VS Code, Zed, Windsurf, and Claude Desktop. Explain something to one agent, and they all know it.
• ✓A knowledge graph, not a pile of notes — entities and relationships with temporal edges, included on every plan.
• ✓Scope isolation — personal, team, and per-agent scopes so client A's context never surfaces in client B's project.