Usage
Options
Initial prompt to send to Claude when the session starts
Permission handling mode:
acceptEdits- Automatically accept all edit operationsbypassPermissions- Bypass all permission checks entirelyplan- Planning mode (Claude creates plans before executing)default- Normal permission prompting behavior
Comma-separated whitelist of allowed tools (e.g.,
Read,Write,Bash)Comma-separated blacklist of disallowed tools
Working directory for Claude (defaults to current directory)
Examples
Permission Modes
- acceptEdits
- bypassPermissions
- plan
- default
Automatically accepts all file edit operations. Claude can still ask questions via the dashboard, but file changes happen without prompts.Use case: Trusted environments where you want speed over control
Tool Filtering
Control which tools Claude can use:Whitelist (allowed-tools)
mcp__omnara__approve) can be used. All others are blocked.
Blacklist (disallowed-tools)
Combined
The
mcp__omnara__approve tool is always allowed, enabling dashboard interaction even with strict tool filtering.How It Works
- Initialization: Creates an agent instance and connects to Omnara MCP server
- Initial Prompt: Sends your specified prompt to Claude
- Background Processing: Claude processes the request and streams responses to dashboard
- Message Queuing: When you respond via dashboard, messages are queued and delivered to Claude
- Session Persistence: Session continues until completed or explicitly ended
Dashboard Interaction
While Claude runs headless, you can interact via dashboard:- Monitor Progress: See Claude’s actions in real-time
- Answer Questions: Respond when Claude asks for input
- View Changes: See git diffs of file modifications
- Control Session: End session or provide new instructions
Logs
Logs are stored at:Use Cases
Automated Code Review
Background Refactoring
CI/CD Integration
Mobile-controlled Development
Start from your computer, control from your phone while commuting