The Session Memory Protocol gives your AI collaborator persistent memory across conversations using @MythOS daily memos. No CLAUDE.md hacks, no manual file management — just structured, cross-client continuity that works on desktop, web, and mobile.

The Problem

Every Claude session starts from zero. Your AI doesn't remember what you worked on yesterday, what decisions you made, or what's still pending. The common workaround: maintain a CLAUDE.md file with session logging instructions that tell the AI to read and write daily notes. This works until it doesn't — context window compression drops the instructions, switching clients loses the chain, and the file drifts stale because there's no system keeping it current.

How It Works

MythOS ships a built-in session_memory prompt via @MythOS MCP that handles the entire protocol automatically.

Session Start

When you invoke the prompt, it:

1. Loads your @Augmentation System — your AI knows your formatting rules, collaboration values, and what prior sessions have learned
2. Reads today's daily memo — previous sessions' timeline entries, open tasks, and notes are immediately in context
3. Your AI knows what happened earlier today, what's pending, and who you are

During the Session

• Key decisions, discoveries, and milestones are logged to the daily memo's Timeline section via add_daily_entry
• New action items are captured as checklist tasks via add_daily_task
• Cross-session references link to relevant memos via @mentions

Session End

• A brief summary is added to the Timeline
• Unfinished work becomes tasks for the next session
• Learnings that should persist beyond today are saved to your Mythos Memory augmentation memo — so every future session, across every client, starts with that knowledge

Setup

Claude.ai / Claude Mobile (OAuth)

If you've already connected MythOS as a connector in Claude, the session_memory prompt is available automatically. Start any conversation with: "Use the session_memory prompt for today's date."

@Claude Code (CLI)

1. Connect MythOS MCP if you haven't already:

claude mcp add mythos \
  --scope user \
  https://mythos.one/api/mcp \
  --header "Accept: application/json, text/event-stream" \
  --header "x-mythos-key: YOUR_API_KEY" \
  --header "x-mythos-username: YOUR_USERNAME"

2. Generate your API key at Settings > API (requires Scholar or Oracle tier).
3. At session start, ask Claude to use the session memory protocol. It will call read_daily_memo and begin logging.

Claude Desktop / Cursor / Windsurf

Add to your MCP config:

{
  "mcpServers": {
    "mythos": {
      "command": "npx",
      "args": ["-y", "mythos-mcp"],
      "env": {
        "MYTHOS_API_KEY": "YOUR_API_KEY",
        "MYTHOS_USERNAME": "YOUR_USERNAME"
      }
    }
  }
}

The session_memory prompt will appear in your client's prompt list.

Why This Beats CLAUDE.md

What Gets Logged

Your daily memo becomes a structured session log:

# Tasks
- [ ] Review PR for auth refactor
- [x] Fix the deploy script timeout
- [ ] Write tests for the import command

# Timeline
* 2:15 PM — Started session. Picked up from yesterday's auth work.
* 2:30 PM — Fixed deploy timeout by increasing the health check interval.
* 3:00 PM — Reviewed import command, found edge case with nested wikilinks.
* 3:45 PM — Session ended. Auth PR still needs review. Import edge case logged as task.

Tomorrow's session reads this and picks up exactly where you left off. This is the feature that makes people stop maintaining CLAUDE.md files. Not because it's more powerful (though it is) — because it's automatic. The daily memo just accumulates context, and every new session starts by reading it. No discipline required. No instructions to maintain. The protocol is the product.

Contexts

• #model-context-protocol
• #mythos-feature