AI agent context persistence that survives every session

Local rules files keep one chat, on one machine, briefed. ReadmeDB gives your agents a shared project brain — markdown files they read and write via MCP so context carries across Cursor, Claude Code, Windsurf, and tomorrow's session, on any machine.

Why chat memory is not enough

Chat transcripts bloat context windows. Vector memory APIs store opaque facts you cannot diff or review. A project brain is different: human-readable files with version history, searchable in one tool call, shared across every MCP host.

What to persist

• decisions.md — architecture choices and rationale
• tasks.md — open work and blockers
• CLAUDE.md / AGENTS.md — conventions agents must follow
• Distilled facts via add_memory → promote when durable

30-second setup

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

Call bootstrap_namespace on first connect to seed starter files. Call get_namespace_summary at session start so agents know what changed.