Installation scope question.

[madanyang]

First thanks for the great software, it has been a life saver many times, which I have been using for single project hence scope was project.

What is your recommendation, install with scope user so memory-keeper is available for all projects or keep using it per project?

[mkreyman]

Glad you found it helpful!

I launch Claude Code from my ~/workspace directory, which contains all my repositories. So that I can access and reference any of them during my coding session. Consequently, my installed MCP servers are not repo-specific.

What I do to isolate saved keys and checkpoints in memory-keeper is I specify a channel (which is usually the repository name).

So I would say something like the following at the end of my session, or whenever I need to,

"Save your context in memory-keeper, the repo_name channel. If you create more than one key, also create a checkpoint to serve as an index for all keys you created".

Something like that... Consequently, at the start of my next session, I would say, "Yesterday we worked on such and such repo. Retrieve relevant context from memory-keeper, the repo_name channel."

Makes sense?

[madanyang]

@mkreyman, thanks for the insights. Unfortunately I am all over the places as I organized my work and personal as separate workspaces.

What do you think about using the hooks at session starts to begin with relevant context from memory-keeper. Is that a feasible option to imply?

[mkreyman]

You could do something like this in your ~/workspace/.claude/settings.local.json:

  {
  "hooks": {
    "SessionStart": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "bash -c 'PROJECT_NAME=$(basename \"$CLAUDE_PROJECT_DIR\" 2>/dev/null || echo \"workspace\"); cat <<EOF\n======================== SESSION START INSTRUCTIONS ========================\nIMPORTANT: Memory-keeper context is available for this workspace.\nCurrent project detected: $PROJECT_NAME\n\nWhen the user sends their first message:\n- If it looks like a channel/project name (single word, kebab-case), treat it as the memory-keeper channel to load\n- Otherwise, use \"$PROJECT_NAME\" as the default channel\n- After loading, briefly confirm: \"📚 Context loaded from memory-keeper ($CHANNEL_NAME)\"\n\nMemory-keeper commands to use:\n- mcp__memory-keeper__context_get({ key: \"CRITICAL_ALWAYS_READ_FIRST_PRINCIPLES\", channel: \"<channel>\" })\n- mcp__memory-keeper__context_get({ priorities: [\"high\"], channel: \"<channel>\" })\n- mcp__memory-keeper__context_get({ category: \"task\", channel: \"<channel>\" })\n=============================================================================\nEOF\n'"
          }
        ]
      }
    ]
  }
  }

User starts a new session and types:

• "repo-name-here" → Claude loads from that channel
• "another-repo-here" → Claude loads from that channel
• "Can you help me with..." → Claude uses current directory name as channel name

[mkreyman]

It won't read from memory-keeper until you say first.

[madanyang]

Thanks will try this