Notes on memory, timelines, context and multi-agent work with Claude Code, Cursor and VS Code.
The Team plan replicates memories, notes, the timeline and chat between teammates as ciphertext we cannot open. What the key is and who actually holds it.
Step-by-step: from pasting files to CLAUDE.md to a memory the agent writes and reads back. What each approach gives Claude Code, and what it leaves out.
MCP lets Claude Code call external tools. How it works, what a shared memory server adds when several agents work on one project, and how to set one up.
What to write in your CLAUDE.md, what to skip, and what to leave to a memory the agent writes for itself and to a status block kept up to date for you.
Context windows have limits, and between sessions an agent starts over. What happens with Claude Code or Cursor on a big project, and what memory changes.