Claude Code Settings Reference: Every Config Option Explained

Core Settings

{
  "defaultModel": "sonnet-4-6",  // Model used by default. Options: opus-4.7, sonnet-4-6, haiku
  "autoMode": false,  // When true, Claude Code acts autonomously without asking for confirmation
  "maxTokens": 8192,  // Max tokens per response. Higher = longer responses, more context
  "temperature": 0.7,  // Response creativity. 0 = deterministic, 1 = creative. Leave at 0.7 for most tasks
  "outputFormat": "text"  // Output format: text, markdown, json
}

MCP Servers

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["@anthropic-ai/mcp-server-github"],
      "env": { "GITHUB_TOKEN": "your-token" }
    }
  }
}

List of MCP server configurations. Each server has its own command, arguments, and environment variables.

Hooks

{
  "hooks": {
    "preCommand": "./scripts/pre-command.sh",  // Runs before each command Claude Code executes
    "postCommit": "echo 'Done'",  // Runs after git commits
    "preSession": null,  // Runs before session starts
    "postSession": null  // Runs after session ends
  }
}

Permissions

{
  "permissions": {
    "allow": ["./src/**"],  // Glob patterns for allowed file access
    "deny": ["./.env"],  // Patterns to never access
    "allowCommand": ["npm", "git", "pytest"]  // Commands Claude Code is allowed to run
  }
}

Session Behavior

{
  "resumeSessions": true,  // Allow resuming previous sessions
  "sessionHistoryLimit": 50,  // Max sessions to keep in history
  "confirmDeletes": true,  // Ask before deleting files
  "confirmGitReset": true,  // Ask before git reset --hard
  "showPlanBeforeActing": true  // Always show plan before executing (can be overridden per request)
}

Developer Options

{
  "verbose": false,  // Extra debug logging
  "debug": false,  // Write debug logs to claude-debug.log
  "telemetry": true  // Anonymous usage telemetry (disable for privacy-sensitive environments)
}