Private AI supports multiple AI coding agents. This guide covers configuring Claude Code, the recommended agent for full feature support.

​
For Users

​
Project-Level Settings

Users can configure Claude Code settings for specific projects via .claude/settings.json in the project directory.
Claude settings file showing project configuration
For detailed Claude Code settings, refer to the official documentation.

​
For Administrators

​
Global Claude Code Settings

Administrators manage global Claude Code settings via ~/.claude/settings.json. This applies to all sessions across the system.

​
Private AI Agent Configuration

Configure how Private AI uses Claude Code for task execution.
Agent configuration page with Claude Code settings

​
Configuration Fields

FieldDescription
ModelClaude model to use (e.g., claude-opus-4-7, claude-sonnet-4-6)
PlanEnable planning mode for complex tasks
Dangerously Skip PermissionsSkip permission prompts (use with caution)
Base Command OverrideCustom command to launch Claude Code
Additional ParametersExtra CLI flags passed to the agent
Environment VariablesKey-value pairs set before execution
Dangerously Skip Permissions disables all safety checks. Only enable in trusted, isolated environments.

​
Private AI MCP Server

Private AI includes a built-in MCP server for Claude Code integration. This server:
  • Authenticates local users via session tokens
  • Provides task management capabilities to Claude Code
The Private AI MCP server is only supported by Claude Code. Other agents (Cursor, Gemini CLI, etc.) cannot use this integration.
When configured, Private AI creates a .claude/vibe-kanban.mcp.json file in the worktree with session credentials.
The vibe-kanban.mcp.json file contains session_id and must never be committed to version control. It is excluded by .gitignore by default.

​
Troubleshooting

  • Verify Claude Code is installed and accessible
  • Check model name is valid
  • Ensure API credentials are configured in Claude Code
  • Confirm using Claude Code (other agents unsupported)
  • Check base_command_override is configured correctly
  • Verify session is active in Private AI
  • Use Plan mode for structured execution
  • Consider Dangerously Skip Permissions only in isolated environments