Separate configuration from conversation history in ~/.claude.json

Feature Request: Separate Configuration from Conversation History in ~/.claude.json

Problem

Currently, ~/.claude.json mixes two very different types of data:

1. Configuration data (MCP servers, settings, user preferences)
2. Conversation history (every prompt entered across all projects)

This causes several issues:

1. File Size and Performance

• The file grows indefinitely with conversation history
• My ~/.claude.json is currently 1300+ lines with history from multiple projects
• Large file size impacts:
• Startup time (parsing large JSON on every launch)
• Write performance (updating the entire file frequently)
• Memory usage

2. Mixed Concerns

{
  "mcpServers": { /* actual config */ },
  "projects": {
    "/some/path": {
      "history": [
        /* 100+ conversation entries */
      ]
    }
  }
}

Configuration (MCP servers, allowed tools) is permanent, while history is ephemeral and grows constantly.

3. Difficult to Manage

• Hard to manually edit MCP server configuration when surrounded by hundreds of history entries
• Can't easily version control just the configuration
• Risk of corruption increases with file size
• No easy way to clear history without losing configuration

4. Privacy Concerns

• All conversation history stored in plain text
• Includes potentially sensitive commands, file paths, and prompts
• Difficult to selectively clean up

Proposed Solution

Split into separate files:

~/.claude/
├── config.json          # MCP servers, global settings
├── sessions.db          # Conversation history (already exists)
└── project-metadata.json # Per-project stats, last used, etc.

config.json

{
  "mcpServers": { /* ... */ },
  "numStartups": 51,
  "autoUpdates": false,
  "forceLoginMethod": "...",
  "subscriptionInfo": { /* ... */ }
}

project-metadata.json

{
  "/home/user/project": {
    "lastSessionId": "...",
    "lastCost": 0.15,
    "allowedTools": [],
    "projectOnboardingSeenCount": 4,
    "exampleFiles": [/* ... */]
  }
}

sessions.db

(Already exists - continue using for full conversation storage)

Benefits

1. Performance: Smaller config file = faster startup
2. Maintainability: Easy to edit MCP configuration
3. Version Control: Can safely commit config.json to dotfiles
4. Privacy: History is in database, easier to manage/purge
5. Stability: Less risk of corrupting important configuration

Migration Path

• Automatically migrate existing ~/.claude.json on first launch
• Keep reading old format for backwards compatibility
• Warn users after migration completes

Alternative: At Minimum

If full separation is too complex, at least:

1. Limit history entries per project (e.g., keep only last 50)
2. Add claude config clear-history command
3. Move history to a separate section that can be easily deleted

---

Current Workaround

Users currently have no good options except:

• Manually editing the large JSON file (risky)
• Periodically deleting the entire file (loses all configuration)
• Living with the growing file size

Related

This aligns with the existing pattern where full conversations are stored in ~/.claude/sessions.db, so history in the JSON file is redundant for anything except recent command autocomplete.